Agents workspace: roster, runs, and a readable run log; layout now eight workspaces

.pi/memory-vault/packs/paperclip-harness/20-context/home.md

@@ -205,3 +205,107 @@ Theme/styling conventions:
 - action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Built Work workspace and polished Board styling (4e3c955)]]
 - session: [[packs/paperclip-harness/80-sessions/rich-discovery-config-2026-11-10-03-01-16|Rich discovery: .config 2026-11-10 03:01:16]]
 - context: [[packs/paperclip-harness/20-context/tmp|.config]]
+
+
+---
+
+## Update (2026-11-10 05:41:48)
+
+## Rich discovery update (2026-11-10 05:41:48)
+
+Reverted. Seven workspaces, one pane each, all drawing again — that is the state you had, restored.
+
+I was wrong to split it. You said panels were missing and I read that as "make panes", then made nineteen. That was my invention, not your request.
+
+## ocr is installed — it has already blocked a merge
+
+```
+/home/sam/.local/bin/ocr
+open-code-review v1.13.6
+```
+
+It is the review gate in `bin/merge`. On `AUD-165` it reported a **critical** finding — `rm -rf "$HOME/tmp"` empties `/tmp` when `HOME` is unset — and the merge was refused; `cleanup.sh` never reached `main`. The Repair screen once said it was missing because a pane started by TUIOS gets a smaller PATH than your shell; that check now looks in `~/.local/bin` and the system profiles as well.
+
+So the review is real, and the interface lacks a place to show the verdict. That is item 4 in the plan.
+
+## Where you see the Pi agents: nowhere yet, and that is the real gap
+
+Paperclip has 281 runs recorded on this instance, each with a 300 KB log I can read:
+
+```
+GET /api/companies/{id}/heartbeat-runs        → 281 runs
+GET /api/heartbeat-runs/{runId}/log           → the agent's own output, grows while it runs
+POST /api/agents/{id}/pause | /resume | /wakeup
+PATCH /api/agents/{id}  {role, adapterConfig (model), runtimeConfig, budgetMonthlyCents}
+```
+
+No screen shows any of it. That is the missing workspace, and it is next.
+
+## The plan, in Paperclip's terms
+
+`docs/PLAN-INTERFACE.md` now holds it, with a table of every Paperclip object and the screen that shows it. The parts that do not exist:
+
+**Agents>agents** — one pane, tabs `roster | runs | log`:
+- roster: every agent with its state, current task, spend and budget
+- runs: the company's runs — agent, task, state, what woke it, duration, cost
+- log: the tail of the chosen run, which is the closest thing to watching Pi work
+
+Actions: `enter` opens a run's log, `p`/`P` pause and resume an agent, `w` wakes one.
+
+**Agents>settings** — role, job title, who it reports to, model, heartbeat, monthly budget. These map straight onto `PATCH /api/agents/{id}`. Two rules: a model change applies to that agent's next run, never one in flight; a budget change is recorded with who made it.
+
+**Tool panels** — Lazygit and Hunk are installed (`/run/current-system/sw/bin/lazygit`, `/etc/profiles/per-user/sam/bin/hunk`) and are **not** wired in. Proposal: `t` in `Work>work` opens a chooser — `shell · lazygit · hunk · git log · ocr` — as a pane in the current workspace, `esc` closes it and focus returns. Nothing is created until you ask.
+
+**Review panel** in `Work>work` — the last verdict and its findings, blocking ones in red, beside the diff. The merge stays a command, because a merge is a decision with a lock, not a keystroke.
+
+## Mouse-1 and Switch
+
+Measured: **not installed** — no `mouse`, `mouse1` or `switch` binary here, and nothing serving them locally. So nothing can be built on them yet.
+
+Where they would fit honestly: cheap, repeatable judgement — "is this task ready", "does this comment answer the question", "which agent should take this" — not prose. That comes after agents, settings, tool panels and the review panel, and only once they are installed and measured against the model already configured for your agents (`omni/deepseek/deepseek-v4-flash`).
+
+## Order
+
+1. Agents>agents (read-only, cannot break anything)
+2. Agents>settings (role, model, budget, pause/resume)
+3. Tool panels (`t` → Lazygit, Hunk, shell, ocr)
+4. Review panel in Work>work
+5. Mouse-1 / Switch, if installed
+
+Each is one pane, one program, and none of them changes the layout you have now.
+
+Committed as `7f18c54`. Say which item you want first and I will build only that.
+
+## Source paths observed
+
+- `.pi/memory-vault/packs/paperclip-harness/20-context/tmp.md`
+- `.pi/memory-vault/packs/paperclip-harness/60-observations/paperclip-harness-tui-editor-pitfall-illegal-rune-literal-af.md`
+- `.pi/memory-vault/packs/paperclip-harness/50-decisions/use-real-tuios-panes-per-workspace-panels-as-panes.md`
+- `.pi/memory-vault/packs/paperclip-harness/80-sessions/paperclip-harness-tui-workspace-rename-new-panels-commit-264.md`
+- `.pi/memory-vault/packs/paperclip-harness/80-sessions/rich-discovery-how-to-tell-whether-the-icons-loaded-2026-11-.md`
+- `home/sam/.local/bin/ocr`
+- `api/agents/me`
+- `api/agents/me/connections`
+- `api/agents/me/inbox-lite`
+- `api/agents/me/inbox/mine`
+- `api/agents/me/secret-proposals`
+- `api/agents/me/secrets`
+- `home/sam/paseo/projects/paperclip-harness/docs/PLAN-INTERFACE.md`
+- `run/current-system/sw/bin/lazygit`
+- `etc/profiles/per-user/sam/bin/hunk`
+- `omni/deepseek/deepseek-v4-flash`
+
+## Evidence
+
+- Confidence: 84%
+- Reason: Update component/repository context with detailed discovered structure and configuration
+
+## Related learned memories <!-- same-turn-2026-11-10 05:41:48 -->
+
+- context: [[packs/paperclip-harness/20-context/paperclip-harness|paperclip-harness]]
+- observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|paperclip-harness runtime caveats: ocr PATH in TUIOS panes, missing mouse/switch, installed tool paths]]
+- decision: [[packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li|Keep 7 single-pane workspaces; map Paperclip objects to screens instead of splitting panes]]
+- action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Reverted 19-pane workspace split and wrote docs/PLAN-INTERFACE.md mapping Paperclip objects to screens]]
+- session: [[packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover|Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap]]
+- session: [[packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48|Rich discovery: home 2026-11-10 05:41:48]]
+- context: [[packs/paperclip-harness/20-context/home|home]]

.pi/memory-vault/packs/paperclip-harness/20-context/paperclip-harness.md

@@ -137,3 +137,44 @@ Committed as `351cb35`. Next, when you are back: tell me which of the six you wa
 - session: [[packs/paperclip-harness/80-sessions/paperclip-harness-tui-six-workspace-build-screens-and-bug-fi|paperclip-harness TUI: six-workspace build, screens, and bug fixes (351cb35)]]
 - session: [[packs/paperclip-harness/80-sessions/rich-discovery-what-you-will-see-when-you-open-it-2026-11-10|Rich discovery: what-you-will-see-when-you-open-it 2026-11-10 03:29:19]]
 - context: [[packs/paperclip-harness/20-context/paperclip-harness|what-you-will-see-when-you-open-it]]
+
+
+---
+
+## Update (2026-11-10 05:41:48)
+
+Repo: `/home/sam/paseo/projects/paperclip-harness` — a TUI harness that maps Paperclip company objects (agents, runs, tasks, workspace) to terminal workspaces/panes.
+
+## Current layout (as of commit 7f18c54)
+- 7 workspaces, one pane per workspace, all drawing: `board`, `work`, `reply`, `new`, `project`, `repair`, `shell`.
+- Board headers look like: `Board · spike-2026-11-10`; work: `Work · AUD-161 Tree render check (no work)`.
+- Do NOT split workspaces into multiple panes unless explicitly asked — a 19-pane split (commit 32fa275) was reverted by the user context.
+
+## Key support files
+- `bin/merge` — merge command with a lock; also the review gate invoking `ocr`.
+- `docs/PLAN-INTERFACE.md` — the design doc mapping Paperclip objects to screens (Agents>agents, Agents>settings, Tool panels, Review panel, Mouse-1/Switch). Maintain this when the interface evolves.
+- `.pi/memory-vault/packs/paperclip-harness/` — pack with 20-context, 50-decisions, 60-observations, 80-sessions subfolders.
+
+## Paperclip API surface used
+- Agents: `PATCH /api/agents/{id}` (role/title/reportsTo/adapterConfig/runtimeConfig/budgetMonthlyCents/status etc.), `POST /api/agents/{id}/pause|resume|wakeup|approve|clear-error`, `GET /api/agents/{id}/configuration|runtime-state|task-sessions|config-revisions|skills`, `POST /api/agents/{id}/skills/sync`.
+- Companies: `GET /api/companies/{companyId}/agents`, `/built-in-agents`, `/heartbeat-runs`.
+- Runs: `GET /api/heartbeat-runs/{runId}/log` (streaming, ~300 KB per run), `/events`.
+- Instance sample: 281 runs recorded. Agent adapter model `omni/deepseek/deepseek-v4-flash`.
+
+## Tooling caveat
+TUIOS panes get a smaller PATH than the user shell — health checks should look in `~/.local/bin` and system profiles, not just PATH.
+
+## Evidence
+
+- Confidence: 90%
+- Reason: Durable repo/component overview; would update an existing context note rather than create transcript noise.
+
+## Related learned memories <!-- same-turn-2026-11-10 05:41:48 -->
+
+- context: [[packs/paperclip-harness/20-context/paperclip-harness|paperclip-harness]]
+- observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|paperclip-harness runtime caveats: ocr PATH in TUIOS panes, missing mouse/switch, installed tool paths]]
+- decision: [[packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li|Keep 7 single-pane workspaces; map Paperclip objects to screens instead of splitting panes]]
+- action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Reverted 19-pane workspace split and wrote docs/PLAN-INTERFACE.md mapping Paperclip objects to screens]]
+- session: [[packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover|Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap]]
+- session: [[packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48|Rich discovery: home 2026-11-10 05:41:48]]
+- context: [[packs/paperclip-harness/20-context/home|home]]

.pi/memory-vault/packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups.md

@@ -248,3 +248,35 @@ Deliverables: built six (plus shell = seven) styled TUI workspaces in the paperc
 - session: [[packs/paperclip-harness/80-sessions/paperclip-harness-tui-six-workspace-build-screens-and-bug-fi|paperclip-harness TUI: six-workspace build, screens, and bug fixes (351cb35)]]
 - session: [[packs/paperclip-harness/80-sessions/rich-discovery-what-you-will-see-when-you-open-it-2026-11-10|Rich discovery: what-you-will-see-when-you-open-it 2026-11-10 03:29:19]]
 - context: [[packs/paperclip-harness/20-context/paperclip-harness|what-you-will-see-when-you-open-it]]
+
+
+---
+
+## Update (2026-11-10 05:41:48)
+
+Deliverable: (1) Reverted commit 32fa275 ("Every workspace split into real panes: 19 panes across 7 workspaces") via 851a15b, restoring the 7-workspace / 1-pane-per-workspace layout the user already had (board, work, reply, new, project, repair, shell). No new layout was invented. (2) Wrote `docs/PLAN-INTERFACE.md` (~7.8 KB) mapping every Paperclip object to a screen, listing missing screens and an order. Commit 7f18c54.
+
+Plan highlights (order):
+1. Agents>agents — roster | runs | log tabs; actions enter (open run log), p/P (pause/resume), w (wake). Read-only first.
+2. Agents>settings — role, job title, reportsTo, model, heartbeat, monthly budget, mapping to PATCH /api/agents/{id}. Model change applies to next run not in-flight; budget changes recorded with actor.
+3. Tool panels — `t` in Work>work opens chooser `shell · lazygit · hunk · git log · ocr` as a pane in the current workspace, esc closes and returns focus. Nothing created until user asks.
+4. Review panel in Work>work — last ocr verdict/findings (blocking in red) beside diff; merges remain a command.
+5. Mouse-1 / Switch only if installed and measured against omni/deepseek/deepseek-v4-flash.
+
+Validation: workspace listing confirmed 7 workspaces with 1 pane each and all drawing.
+Follow-up: user asked to pick item 1..5; only build the chosen item.
+
+## Evidence
+
+- Confidence: 90%
+- Reason: Completed concrete work with commit hashes, deliverables, and an explicit follow-up ask.
+
+## Related learned memories <!-- same-turn-2026-11-10 05:41:48 -->
+
+- context: [[packs/paperclip-harness/20-context/paperclip-harness|paperclip-harness]]
+- observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|paperclip-harness runtime caveats: ocr PATH in TUIOS panes, missing mouse/switch, installed tool paths]]
+- decision: [[packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li|Keep 7 single-pane workspaces; map Paperclip objects to screens instead of splitting panes]]
+- action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Reverted 19-pane workspace split and wrote docs/PLAN-INTERFACE.md mapping Paperclip objects to screens]]
+- session: [[packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover|Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap]]
+- session: [[packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48|Rich discovery: home 2026-11-10 05:41:48]]
+- context: [[packs/paperclip-harness/20-context/home|home]]

.pi/memory-vault/packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li.md

@@ -82,3 +82,36 @@ Consequences: responsive layouts by terminal width (>=150 wide, 110-149 two pane
 - session: [[packs/paperclip-harness/80-sessions/rich-discovery-etc-2026-11-10-02-26-18|Rich discovery: etc 2026-11-10 02:26:18]]
 - context: [[packs/paperclip-harness/20-context/home|etc]]
 - observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|etc repository structure and configuration patterns]]
+
+
+---
+
+## Update (2026-11-10 05:41:48)
+
+Decision: Do not split workspaces into multiple panes. Keep the existing 7-workspace / 1-pane-per-workspace layout and evolve the interface by adding/mapping screens (Agents>agents, Agents>settings, tool panels, review panel, optional Mouse-1/Switch).
+
+Context: The user said some panels were "missing". The agent interpreted this as "split workspaces into real panes" and created 19 panes across 7 workspaces (commit 32fa275). The user rejected it; it was reverted (851a15b) and replaced by a plan (7f18c54).
+
+Rationale: The user's existing layout should not be replaced by an agent's invention. Missing functionality should be added as focused panes/screens, each one pane / one program, without changing current layout.
+
+Consequences / review caveats:
+- New features are added as one pane per program (e.g. `t` chooser opens lazygit/hunk/shell/ocr as a pane in the current workspace).
+- Prefer read-only screens first (Agents>agents) before mutating settings.
+- Merges stay a command (`bin/merge`), not a keystroke; merges are decisions with a lock.
+- Model changes apply to next run only; budget changes are attributed.
+- Only build the specific item the user asks for next (items 1–5 in PLAN-INTERFACE.md).
+
+## Evidence
+
+- Confidence: 90%
+- Reason: A clear architectural/convention decision with rationale, alternatives (rejected pane split), and consequences.
+
+## Related learned memories <!-- same-turn-2026-11-10 05:41:48 -->
+
+- context: [[packs/paperclip-harness/20-context/paperclip-harness|paperclip-harness]]
+- observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|paperclip-harness runtime caveats: ocr PATH in TUIOS panes, missing mouse/switch, installed tool paths]]
+- decision: [[packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li|Keep 7 single-pane workspaces; map Paperclip objects to screens instead of splitting panes]]
+- action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Reverted 19-pane workspace split and wrote docs/PLAN-INTERFACE.md mapping Paperclip objects to screens]]
+- session: [[packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover|Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap]]
+- session: [[packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48|Rich discovery: home 2026-11-10 05:41:48]]
+- context: [[packs/paperclip-harness/20-context/home|home]]

.pi/memory-vault/packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns.md

@@ -778,3 +778,123 @@ Tell me which of those you want first, and whether the icons appear after reopen
 - session: [[packs/paperclip-harness/80-sessions/paperclip-harness-tui-icon-font-fix-help-filter-unblock-safe|paperclip-harness TUI: icon font fix, help/filter/unblock, safe rebuild (commit 09e32da)]]
 - session: [[packs/paperclip-harness/80-sessions/rich-discovery-config-2026-11-10-05-00-01|Rich discovery: .config 2026-11-10 05:00:01]]
 - context: [[packs/paperclip-harness/20-context/tmp|.config]]
+
+
+---
+
+## Update (2026-11-10 05:41:48)
+
+- TUIOS-spawned panes inherit a smaller PATH than the interactive shell; the Repair health check for `ocr` originally reported it missing. Fixed by also checking `~/.local/bin` and system profiles. ocr lives at `/home/sam/.local/bin/ocr` (open-code-review v1.13.6, sha 891df1b6, built 2026-11-05).
+- ocr is a real merge gate wired into `bin/merge`. Verified: on AUD-165 it reported a critical finding (`rm -rf "$HOME/tmp"` empties /tmp when HOME is unset) and refused the merge, so `cleanup.sh` never reached main.
+- Lazygit: `/run/current-system/sw/bin/lazygit`. Hunk: `/etc/profiles/per-user/sam/bin/hunk`. Both installed, not wired into UI.
+- `mouse`, `mouse1`, `switch` binaries are NOT installed and nothing serves them locally — no Mouse-1/Switch integration is currently possible without installation.
+- Paperclip agent API surface includes pause/resume/wakeup/approve/clear-error endpoints and PATCH /api/agents/{id} fields (role, adapterConfig, runtimeConfig, budgetMonthlyCents, etc.). Runs are discoverable via heartbeat-runs with streaming logs at /api/heartbeat-runs/{runId}/log.
+
+## Evidence
+
+- Confidence: 90%
+- Reason: Factual runtime discoveries with exact paths and a verified merge-blocking example; durable for future work.
+
+
+---
+
+## Update (2026-11-10 05:41:48)
+
+## Structural/configuration observations (2026-11-10 05:41:48)
+
+Reverted. Seven workspaces, one pane each, all drawing again — that is the state you had, restored.
+
+I was wrong to split it. You said panels were missing and I read that as "make panes", then made nineteen. That was my invention, not your request.
+
+## ocr is installed — it has already blocked a merge
+
+```
+/home/sam/.local/bin/ocr
+open-code-review v1.13.6
+```
+
+It is the review gate in `bin/merge`. On `AUD-165` it reported a **critical** finding — `rm -rf "$HOME/tmp"` empties `/tmp` when `HOME` is unset — and the merge was refused; `cleanup.sh` never reached `main`. The Repair screen once said it was missing because a pane started by TUIOS gets a smaller PATH than your shell; that check now looks in `~/.local/bin` and the system profiles as well.
+
+So the review is real, and the interface lacks a place to show the verdict. That is item 4 in the plan.
+
+## Where you see the Pi agents: nowhere yet, and that is the real gap
+
+Paperclip has 281 runs recorded on this instance, each with a 300 KB log I can read:
+
+```
+GET /api/companies/{id}/heartbeat-runs        → 281 runs
+GET /api/heartbeat-runs/{runId}/log           → the agent's own output, grows while it runs
+POST /api/agents/{id}/pause | /resume | /wakeup
+PATCH /api/agents/{id}  {role, adapterConfig (model), runtimeConfig, budgetMonthlyCents}
+```
+
+No screen shows any of it. That is the missing workspace, and it is next.
+
+## The plan, in Paperclip's terms
+
+`docs/PLAN-INTERFACE.md` now holds it, with a table of every Paperclip object and the screen that shows it. The parts that do not exist:
+
+**Agents>agents** — one pane, tabs `roster | runs | log`:
+- roster: every agent with its state, current task, spend and budget
+- runs: the company's runs — agent, task, state, what woke it, duration, cost
+- log: the tail of the chosen run, which is the closest thing to watching Pi work
+
+Actions: `enter` opens a run's log, `p`/`P` pause and resume an agent, `w` wakes one.
+
+**Agents>settings** — role, job title, who it reports to, model, heartbeat, monthly budget. These map straight onto `PATCH /api/agents/{id}`. Two rules: a model change applies to that agent's next run, never one in flight; a budget change is recorded with who made it.
+
+**Tool panels** — Lazygit and Hunk are installed (`/run/current-system/sw/bin/lazygit`, `/etc/profiles/per-user/sam/bin/hunk`) and are **not** wired in. Proposal: `t` in `Work>work` opens a chooser — `shell · lazygit · hunk · git log · ocr` — as a pane in the current workspace, `esc` closes it and focus returns. Nothing is created until you ask.
+
+**Review panel** in `Work>work` — the last verdict and its findings, blocking ones in red, beside the diff. The merge stays a command, because a merge is a decision with a lock, not a keystroke.
+
+## Mouse-1 and Switch
+
+Measured: **not installed** — no `mouse`, `mouse1` or `switch` binary here, and nothing serving them locally. So nothing can be built on them yet.
+
+Where they would fit honestly: cheap, repeatable judgement — "is this task ready", "does this comment answer the question", "which agent should take this" — not prose. That comes after agents, settings, tool panels and the review panel, and only once they are installed and measured against the model already configured for your agents (`omni/deepseek/deepseek-v4-flash`).
+
+## Order
+
+1. Agents>agents (read-only, cannot break anything)
+2. Agents>settings (role, model, budget, pause/resume)
+3. Tool panels (`t` → Lazygit, Hunk, shell, ocr)
+4. Review panel in Work>work
+5. Mouse-1 / Switch, if installed
+
+Each is one pane, one program, and none of them changes the layout you have now.
+
+Committed as `7f18c54`. Say which item you want first and I will build only that.
+
+## Source paths observed
+
+- `.pi/memory-vault/packs/paperclip-harness/20-context/tmp.md`
+- `.pi/memory-vault/packs/paperclip-harness/60-observations/paperclip-harness-tui-editor-pitfall-illegal-rune-literal-af.md`
+- `.pi/memory-vault/packs/paperclip-harness/50-decisions/use-real-tuios-panes-per-workspace-panels-as-panes.md`
+- `.pi/memory-vault/packs/paperclip-harness/80-sessions/paperclip-harness-tui-workspace-rename-new-panels-commit-264.md`
+- `.pi/memory-vault/packs/paperclip-harness/80-sessions/rich-discovery-how-to-tell-whether-the-icons-loaded-2026-11-.md`
+- `home/sam/.local/bin/ocr`
+- `api/agents/me`
+- `api/agents/me/connections`
+- `api/agents/me/inbox-lite`
+- `api/agents/me/inbox/mine`
+- `api/agents/me/secret-proposals`
+- `api/agents/me/secrets`
+- `home/sam/paseo/projects/paperclip-harness/docs/PLAN-INTERFACE.md`
+- `run/current-system/sw/bin/lazygit`
+- `etc/profiles/per-user/sam/bin/hunk`
+- `omni/deepseek/deepseek-v4-flash`
+
+## Evidence
+
+- Confidence: 82%
+- Reason: Captured factual repo structure/configuration patterns from rich discovery
+
+## Related learned memories <!-- same-turn-2026-11-10 05:41:48 -->
+
+- context: [[packs/paperclip-harness/20-context/paperclip-harness|paperclip-harness]]
+- observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|paperclip-harness runtime caveats: ocr PATH in TUIOS panes, missing mouse/switch, installed tool paths]]
+- decision: [[packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li|Keep 7 single-pane workspaces; map Paperclip objects to screens instead of splitting panes]]
+- action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Reverted 19-pane workspace split and wrote docs/PLAN-INTERFACE.md mapping Paperclip objects to screens]]
+- session: [[packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover|Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap]]
+- session: [[packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48|Rich discovery: home 2026-11-10 05:41:48]]
+- context: [[packs/paperclip-harness/20-context/home|home]]

.pi/memory-vault/packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover.md

@@ -0,0 +1,70 @@
+---
+type: session
+id: session.paperclip-harness.paperclip-harness-reverted-pane-split-interface-gap-discover
+title: Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap
+status: active
+source_of_truth: false
+freshness: current
+last_reviewed: 2026-11-10
+tags:
+  - pack/paperclip-harness
+  - agent-memory/session
+  - paperclip-harness
+  - tui
+  - plan-interface
+  - ocr
+  - agents
+  - heartbeat-runs
+  - revert
+  - mouse-1
+---
+
+# Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap
+
+Sanitized discovery snapshot from a paperclip-harness workspace session at /home/sam/paseo/projects/paperclip-harness.
+
+## Repo state change
+A change titled "Every workspace split into real panes: 19 panes across 7 workspaces, one program per panel" (commit 32fa275) was reverted in 851a15b, then a follow-up commit 7f18c54 "Revert the pane split; write the plan that maps every Paperclip object to a screen" was made. Current state: 7 workspaces, one pane each, all drawing.
+- Workspaces: board, work, reply, new, project, repair (plus a Shell workspace = 7 total).
+- Workspace titles: `Board · spike-2026-11-10`, `Work · AUD-161 Tree render check (no work)`, `Reply · ...`, `Author · new task in spike-2026-11-10`, `Project · spike-2026-11-10`, `Repair · layout and health`.
+- Lesson: the user said "panels were missing"; the agent over-interpreted and built 19 panes; correct interpretation was not to split into panes. Reverted.
+
+## Tooling discovered
+- `ocr` (open-code-review) v1.13.6 at `/home/sam/.local/bin/ocr`. It is the review gate in `bin/merge`. On AUD-165 it flagged a **critical** finding (`rm -rf "$HOME/tmp"` empties /tmp when HOME unset) and blocked the merge; cleanup.sh never reached main.
+- Repair screen previously said ocr was missing because a TUIOS pane started with a smaller PATH than the shell; the check now also looks in `~/.local/bin` and system profiles.
+- Lazygit at `/run/current-system/sw/bin/lazygit`, Hunk at `/etc/profiles/per-user/sam/bin/hunk` — installed but not wired into the interface.
+- `mouse`, `mouse1`, `switch` binaries: **not installed**, nothing serving them locally (checked; local listeners on 8080, 3000, 8000 are unrelated).
+
+## Paperclip harness facts
+- 281 runs recorded on this instance.
+- `GET /api/companies/{id}/heartbeat-runs` lists runs (sample keys include agentId, companyId, invocationSource, exitCode, status, error, logRef, store=local_file).
+- `GET /api/heartbeat-runs/{runId}/log` streams the agent's own output (sample ~300 KB log).
+- `GET /api/heartbeat-runs/{runId}/events`.
+- Agent control: `POST /api/agents/{id}/pause | /resume | /wakeup | /approve | /clear-error`; `PATCH /api/agents/{id}` accepts name, role, title, icon, reportsTo, capabilities, desiredSkills, adapterType, adapterConfig, instructionsBundle, runtimeConfig, defaultEnvironmentId, budgetMonthlyCents, metadata, storedSessionId, applyStoredClaudeLogin, replaceAdapterConfig, status, spentMonthlyCents.
+- Read APIs: `GET /api/agents/{id}/configuration`, `/runtime-state`, `/task-sessions`, `/config-revisions`, `/skills`; `POST /api/agents/{id}/skills/sync`; `GET /api/companies/{companyId}/agents`, `/built-in-agents`.
+- Adapter/model currently configured for the user's agents: `omni/deepseek/deepseek-v4-flash` (referenced in the roadmap for Mouse-1/Switch).
+
+## Missing screens / roadmap written by the assistant
+- `docs/PLAN-INTERFACE.md` (written, 7,843 bytes) maps every Paperclip object to a screen and lists missing screens.
+- Identified gap: no screen shows agents or runs — this is the missing workspace.
+- Proposed order: 1) Agents>agents (roster/runs/log, read-only; enter opens run log, p/P pause/resume, w wake), 2) Agents>settings (role, model, heartbeat, budget — maps to PATCH /api/agents/{id}), 3) Tool panels (`t` in Work>work opens chooser shell · lazygit · hunk · git log · ocr as a pane in current workspace), 4) Review panel in Work>work (last ocr verdict/findings), 5) Mouse-1/Switch only if installed.
+- Rules stated: model change applies to agent's next run, never in-flight; budget changes recorded with who made it; merges stay a command, not a keystroke.
+
+## Evidence
+
+- Confidence: 90%
+- Reason: Rich discovery: repo state change, tooling, API surface, and a written roadmap document with concrete commit hashes and paths.
+
+## Related
+
+- [[packs/paperclip-harness/00-system/pi-agent/memory-manifest|Memory Manifest]]
+
+## Related learned memories <!-- same-turn-2026-11-10 05:41:48 -->
+
+- context: [[packs/paperclip-harness/20-context/paperclip-harness|paperclip-harness]]
+- observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|paperclip-harness runtime caveats: ocr PATH in TUIOS panes, missing mouse/switch, installed tool paths]]
+- decision: [[packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li|Keep 7 single-pane workspaces; map Paperclip objects to screens instead of splitting panes]]
+- action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Reverted 19-pane workspace split and wrote docs/PLAN-INTERFACE.md mapping Paperclip objects to screens]]
+- session: [[packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover|Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap]]
+- session: [[packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48|Rich discovery: home 2026-11-10 05:41:48]]
+- context: [[packs/paperclip-harness/20-context/home|home]]

.pi/memory-vault/packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48.md

@@ -0,0 +1,123 @@
+---
+type: session
+id: session.paperclip-harness.rich-discovery-home-2026-11-10-05-41-48
+title: Rich discovery: home 2026-11-10 05:41:48
+status: active
+source_of_truth: false
+freshness: current
+last_reviewed: 2026-11-10
+tags:
+  - pack/paperclip-harness
+  - agent-memory/session
+  - autolearn
+  - rich-discovery
+  - repo/home
+---
+
+# Rich discovery: home 2026-11-10 05:41:48
+
+Sanitized rich discovery snapshot captured automatically from the completed turn.
+
+
+## Source paths observed
+
+- `.pi/memory-vault/packs/paperclip-harness/20-context/tmp.md`
+- `.pi/memory-vault/packs/paperclip-harness/60-observations/paperclip-harness-tui-editor-pitfall-illegal-rune-literal-af.md`
+- `.pi/memory-vault/packs/paperclip-harness/50-decisions/use-real-tuios-panes-per-workspace-panels-as-panes.md`
+- `.pi/memory-vault/packs/paperclip-harness/80-sessions/paperclip-harness-tui-workspace-rename-new-panels-commit-264.md`
+- `.pi/memory-vault/packs/paperclip-harness/80-sessions/rich-discovery-how-to-tell-whether-the-icons-loaded-2026-11-.md`
+- `home/sam/.local/bin/ocr`
+- `api/agents/me`
+- `api/agents/me/connections`
+- `api/agents/me/inbox-lite`
+- `api/agents/me/inbox/mine`
+- `api/agents/me/secret-proposals`
+- `api/agents/me/secrets`
+- `home/sam/paseo/projects/paperclip-harness/docs/PLAN-INTERFACE.md`
+- `run/current-system/sw/bin/lazygit`
+- `etc/profiles/per-user/sam/bin/hunk`
+- `omni/deepseek/deepseek-v4-flash`
+
+## Final answer snapshot
+
+Reverted. Seven workspaces, one pane each, all drawing again — that is the state you had, restored.
+
+I was wrong to split it. You said panels were missing and I read that as "make panes", then made nineteen. That was my invention, not your request.
+
+## ocr is installed — it has already blocked a merge
+
+```
+/home/sam/.local/bin/ocr
+open-code-review v1.13.6
+```
+
+It is the review gate in `bin/merge`. On `AUD-165` it reported a **critical** finding — `rm -rf "$HOME/tmp"` empties `/tmp` when `HOME` is unset — and the merge was refused; `cleanup.sh` never reached `main`. The Repair screen once said it was missing because a pane started by TUIOS gets a smaller PATH than your shell; that check now looks in `~/.local/bin` and the system profiles as well.
+
+So the review is real, and the interface lacks a place to show the verdict. That is item 4 in the plan.
+
+## Where you see the Pi agents: nowhere yet, and that is the real gap
+
+Paperclip has 281 runs recorded on this instance, each with a 300 KB log I can read:
+
+```
+GET /api/companies/{id}/heartbeat-runs        → 281 runs
+GET /api/heartbeat-runs/{runId}/log           → the agent's own output, grows while it runs
+POST /api/agents/{id}/pause | /resume | /wakeup
+PATCH /api/agents/{id}  {role, adapterConfig (model), runtimeConfig, budgetMonthlyCents}
+```
+
+No screen shows any of it. That is the missing workspace, and it is next.
+
+## The plan, in Paperclip's terms
+
+`docs/PLAN-INTERFACE.md` now holds it, with a table of every Paperclip object and the screen that shows it. The parts that do not exist:
+
+**Agents>agents** — one pane, tabs `roster | runs | log`:
+- roster: every agent with its state, current task, spend and budget
+- runs: the company's runs — agent, task, state, what woke it, duration, cost
+- log: the tail of the chosen run, which is the closest thing to watching Pi work
+
+Actions: `enter` opens a run's log, `p`/`P` pause and resume an agent, `w` wakes one.
+
+**Agents>settings** — role, job title, who it reports to, model, heartbeat, monthly budget. These map straight onto `PATCH /api/agents/{id}`. Two rules: a model change applies to that agent's next run, never one in flight; a budget change is recorded with who made it.
+
+**Tool panels** — Lazygit and Hunk are installed (`/run/current-system/sw/bin/lazygit`, `/etc/profiles/per-user/sam/bin/hunk`) and are **not** wired in. Proposal: `t` in `Work>work` opens a chooser — `shell · lazygit · hunk · git log · ocr` — as a pane in the current workspace, `esc` closes it and focus returns. Nothing is created until you ask.
+
+**Review panel** in `Work>work` — the last verdict and its findings, blocking ones in red, beside the diff. The merge stays a command, because a merge is a decision with a lock, not a keystroke.
+
+## Mouse-1 and Switch
+
+Measured: **not installed** — no `mouse`, `mouse1` or `switch` binary here, and nothing serving them locally. So nothing can be built on them yet.
+
+Where they would fit honestly: cheap, repeatable judgement — "is this task ready", "does this comment answer the question", "which agent should take this" — not prose. That comes after agents, settings, tool panels and the review panel, and only once they are installed and measured against the model already configured for your agents (`omni/deepseek/deepseek-v4-flash`).
+
+## Order
+
+1. Agents>agents (read-only, cannot break anything)
+2. Agents>settings (role, model, budget, pause/resume)
+3. Tool panels (`t` → Lazygit, Hunk, shell, ocr)
+4. Review panel in Work>work
+5. Mouse-1 / Switch, if installed
+
+Each is one pane, one program, and none of them changes the layout you have now.
+
+Committed as `7f18c54`. Say which item you want first and I will build only that.
+
+## Evidence
+
+- Confidence: 86%
+- Reason: Preserve full sanitized discovery detail from a rich investigation turn
+
+## Related
+
+- [[packs/paperclip-harness/00-system/pi-agent/memory-manifest|Memory Manifest]]
+
+## Related learned memories <!-- same-turn-2026-11-10 05:41:48 -->
+
+- context: [[packs/paperclip-harness/20-context/paperclip-harness|paperclip-harness]]
+- observation: [[packs/paperclip-harness/60-observations/tmp-repository-structure-and-configuration-patterns|paperclip-harness runtime caveats: ocr PATH in TUIOS panes, missing mouse/switch, installed tool paths]]
+- decision: [[packs/paperclip-harness/50-decisions/paperclip-harness-ui-six-workspaces-spec-first-four-panel-li|Keep 7 single-pane workspaces; map Paperclip objects to screens instead of splitting panes]]
+- action: [[packs/paperclip-harness/40-actions/2026-11-10-wrote-docs-ui-layout-md-and-three-live-layout-mockups|Reverted 19-pane workspace split and wrote docs/PLAN-INTERFACE.md mapping Paperclip objects to screens]]
+- session: [[packs/paperclip-harness/80-sessions/paperclip-harness-reverted-pane-split-interface-gap-discover|Paperclip-harness — reverted pane split, interface gap discovery, and PLAN-INTERFACE roadmap]]
+- session: [[packs/paperclip-harness/80-sessions/rich-discovery-home-2026-11-10-05-41-48|Rich discovery: home 2026-11-10 05:41:48]]
+- context: [[packs/paperclip-harness/20-context/home|home]]

bin/board

@@ -47,6 +47,7 @@ def pages(cfg, folder):
         ("shell", "Shell", "exec zsh -i"),
         ("board", "Board", "%s -screen board -project %s -folder %s" % (shlex.quote(bin_path), proj, shlex.quote(folder))),
         ("work", "Work", "%s -screen work -folder %s" % (shlex.quote(bin_path), shlex.quote(folder))),
+        ("agents", "Agents", "%s -screen agents -folder %s" % (shlex.quote(bin_path), shlex.quote(folder))),
         ("reply", "Reply", "%s -screen reply -folder %s" % (shlex.quote(bin_path), shlex.quote(folder))),
         ("new", "New task", "%s -screen new -project %s" % (shlex.quote(bin_path), proj)),
         ("project", "Project", "%s -screen project -folder %s" % (shlex.quote(bin_path), shlex.quote(folder))),

docs/WORKSPACES.md

@@ -0,0 +1,96 @@
+# Every workspace and pane, at the end
+
+The finished shape, so a gap is visible at a glance. One pane per workspace unless stated;
+panels are what the pane's program draws, switched with Tab or the tab keys.
+
+Status: **built** means running in the project session now. **missing** means planned, not
+written yet.
+
+## 1. Shell
+
+| Pane | Shows | Status |
+|---|---|---|
+| `Shell>shell` | a prompt in the project folder | built |
+
+## 2. Board
+
+| Pane | Shows | Status |
+|---|---|---|
+| `Board>tasks` | tasks grouped todo, running, blocked, review, done; each row: identifier, state, assignee, title, what it waits on | built |
+| `Board>detail` | the chosen task: state, assignee, blockers, verify command, last run, cost | built |
+| `Board>needs` | decisions waiting on a person; second tab: the chosen task's activity | built |
+
+## 3. Work
+
+| Pane | Shows | Status |
+|---|---|---|
+| `Work>runs` | run timeline and activity, with cost per run | built |
+| `Work>change` | worktree, branch, base, commits, changed files, the diff | built |
+| `Work>facts` | task, state, worktree path, branch, base, cost | built |
+| `Work>review` | the last ocr verdict for this task: findings, blocking ones in red, and whether a merge would be refused | **missing** |
+
+## 4. Reply
+
+| Pane | Shows | Status |
+|---|---|---|
+| `Reply>task` | the task above the conversation: identifier, state, assignee, verify command | built |
+| `Reply>thread` | every message, with author and time | built |
+| `Reply>replybox` | the reply box, and the state that will be sent with it | built |
+
+## 5. New task
+
+| Pane | Shows | Status |
+|---|---|---|
+| `New task>form` | title, what it must do, what must finish first, the command that proves it is done, whether to assign | built |
+| `New task>guidance` | what a task needs: one thing, a command, a boundary, a blocker | built |
+| `New task>preview` | exactly the description the agent will receive | built |
+
+## 6. Agents — the missing workspace
+
+| Pane | Shows | Status |
+|---|---|---|
+| `Agents>roster` | every agent: name, state, spend this month, budget | built |
+| `Agents>runs` | the company's runs, newest first: agent, task, state, what woke it, duration | built |
+| `Agents>log` | the chosen run's output, deltas gathered back into paragraphs — this is watching Pi work | built |
+| `Agents>settings` | role, job title, who it reports to, model, heartbeat, monthly budget | **missing** |
+
+## 7. Project
+
+| Pane | Shows | Status |
+|---|---|---|
+| `Project>project` | project, folder, machine, session, agents | built |
+| `Project>costs` | spend per task and the total | built |
+| `Project>settings` | the project's own file, and the limits Paperclip holds | built |
+
+## 8. Repair
+
+| Pane | Shows | Status |
+|---|---|---|
+| `Repair>repair` | health: three-state checks and an icon test row | built |
+| `Repair>layout` | what TUIOS actually has, read live | built |
+| `Repair>rebuild-log` | the last rebuild's output | built |
+
+## Tool panes — any of these, in any workspace, on demand
+
+A tool pane is not part of a layout. `t` opens a chooser in the workspace you are in, the
+tool appears beside what you were reading, `esc` closes it and focus returns. This is how
+Yazi, fzf, Neovim, Lazygit, Hunk and anything else fits without being rebuilt by us.
+
+| Tool | What it is for | Status |
+|---|---|---|
+| Lazygit | branches, commits, staging, history | installed, **not wired** |
+| Hunk | reading a change as a review | installed, **not wired** |
+| Yazi | files | installed, **not wired** |
+| Neovim | editing a file the task touched | installed, **not wired** |
+| fzf | picking anything | installed, **not wired** |
+| shell | any command the interface does not cover | built (`Shell>shell`) |
+| ocr | the review of the current change | installed, behind `bin/review` |
+
+## Count
+
+- **21 panels built**, across 8 workspaces.
+- **2 panels missing**: `Work>review` and `Agents>settings`.
+- **6 tool panes** to wire: Lazygit, Hunk, Yazi, Neovim, fzf, ocr.
+
+Nothing in the missing list changes the layout: each is one more panel inside a pane that
+already exists, or one pane-opener for a tool.

tui/agents.go

@@ -0,0 +1,633 @@
+package main
+
+// The Agents workspace: who is working, on what, and for how much.
+//
+// This is the view the project was missing. Paperclip already records every
+// agent, every run and every run's output; nothing here is invented, and
+// nothing here is stored. Three panels in one pane, switched with tabs:
+//
+//   roster   every agent, its state, its spend and its budget
+//   runs     every run in the company, newest first
+//   log      the output of the chosen run — watching Pi work
+//
+// Pause, resume and wake are the only writes, and each one is a single call to
+// Paperclip, which owns the agent.
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+	"time"
+
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+
+	"paperclip-harness/tui/paperclip"
+)
+
+type agentsMsg struct {
+	agents []paperclip.Agent
+	names  map[string]string
+	err    error
+}
+
+// runsMsg is separate from agentsMsg on purpose: the company's run list is over
+// a megabyte and takes eight seconds, while the roster is small and instant.
+// Reading them together made the whole screen wait for the slow one.
+type runsMsg struct {
+	runs []paperclip.AgentRun
+	err  error
+}
+
+type runLogMsg struct {
+	runID string
+	text  string
+	err   error
+}
+
+type agentsModel struct {
+	client *paperclip.Client
+	folder string
+
+	agents   []paperclip.Agent
+	runs     []paperclip.AgentRun
+	names    map[string]string
+	logRunID string
+	logText  string
+	err      error
+	notice   string
+	readAt   time.Time
+
+	tab      int
+	cursor   int
+	scroll   int
+	showHelp bool
+
+	width, height int
+}
+
+var agentTabs = []string{"roster", "runs", "log"}
+
+func newAgentsModel(c *paperclip.Client, folder string) agentsModel {
+	return agentsModel{client: c, folder: folder}
+}
+
+func (m agentsModel) Init() tea.Cmd {
+	return tea.Batch(m.load(), tickEvery(5*time.Second))
+}
+
+func (m agentsModel) load() tea.Cmd {
+	c := m.client
+	return func() tea.Msg {
+		agents, err := c.Agents()
+		if err != nil {
+			return agentsMsg{err: err}
+		}
+		names := map[string]string{}
+		for _, a := range agents {
+			names[a.ID] = a.Name
+		}
+		return agentsMsg{agents: agents, names: names}
+	}
+}
+
+// loadRuns reads the company's runs. Called when the runs tab is opened and
+// when it is asked for, never on the timer: it is a megabyte and eight seconds.
+func (m agentsModel) loadRuns() tea.Cmd {
+	c := m.client
+	return func() tea.Msg {
+		runs, err := c.CompanyRuns()
+		if err != nil {
+			return runsMsg{err: err}
+		}
+		if len(runs) > 200 {
+			runs = runs[:200]
+		}
+		return runsMsg{runs: runs}
+	}
+}
+
+func (m agentsModel) loadLog(runID string) tea.Cmd {
+	c := m.client
+	return func() tea.Msg {
+		if runID == "" {
+			return runLogMsg{}
+		}
+		// The tail, not the whole thing: a run log here is tens of thousands of
+		// lines and only the end says what is happening now.
+		// Read the whole log: the stored form is one JSON envelope, and a
+		// partial read cuts the envelope in half and makes it unreadable.
+		text, err := c.RunLog(runID, 8<<20)
+		return runLogMsg{runID: runID, text: text, err: err}
+	}
+}
+
+func (m agentsModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
+	switch msg := msg.(type) {
+	case tea.WindowSizeMsg:
+		m.width, m.height = msg.Width, msg.Height
+		return m, nil
+
+	case tickMsg:
+		return m, tea.Batch(m.load(), tickEvery(5*time.Second))
+
+	case tea.KeyMsg:
+		if m.showHelp {
+			m.showHelp = false
+			return m, nil
+		}
+		switch msg.String() {
+		case "?":
+			m.showHelp = true
+			return m, nil
+		case "q", "ctrl+c":
+			return m, tea.Quit
+		case "r":
+			switch m.tab {
+			case 1:
+				return m, m.loadRuns()
+			case 2:
+				return m, m.loadLog(m.logRunID)
+			}
+			return m, m.load()
+		case "R":
+			if err := startRebuild(m.folder); err != nil {
+				m.err = err
+				return m, nil
+			}
+			return m, tea.Quit
+		case "[", "h":
+			if m.tab > 0 {
+				m.tab--
+				m.scroll = 0
+			}
+			return m, nil
+		case "]", "l", "tab":
+			if m.tab < len(agentTabs)-1 {
+				m.tab++
+				m.scroll = 0
+				if m.tab == 1 && len(m.runs) == 0 {
+					return m, m.loadRuns()
+				}
+			}
+			return m, nil
+		case "up", "k":
+			m.scroll--
+			return m, nil
+		case "down", "j":
+			m.scroll++
+			return m, nil
+		case "g":
+			m.scroll = 0
+			return m, nil
+		case "G":
+			m.scroll = 1 << 30
+			return m, nil
+		case "enter":
+			if m.tab == 1 {
+				if r := m.chosenRun(); r != nil {
+					m.logRunID = r.ID
+					m.tab = 2
+					m.scroll = 1 << 30
+					return m, m.loadLog(r.ID)
+				}
+			}
+			return m, nil
+		case "p", "P", "w":
+			agent := m.chosenAgent()
+			if agent == nil {
+				return m, nil
+			}
+			c := m.client
+			action := msg.String()
+			return m, func() tea.Msg {
+				var err error
+				switch action {
+				case "p":
+					err = c.PauseAgent(agent.ID)
+				case "P":
+					err = c.ResumeAgent(agent.ID)
+				case "w":
+					err = c.WakeAgent(agent.ID)
+				}
+				if err != nil {
+					return errMsg{err}
+				}
+				agents, err := c.Agents()
+				if err != nil {
+					return errMsg{err}
+				}
+				return agentsMsg{agents: agents}
+			}
+		case "esc":
+			return m, nil
+		}
+
+	case agentsMsg:
+		if msg.agents != nil {
+			m.agents = msg.agents
+		}
+		m.names = msg.names
+		if msg.err != nil {
+			m.err = msg.err
+		} else {
+			m.err = nil
+		}
+		m.readAt = time.Now()
+		return m, nil
+
+	case runsMsg:
+		if msg.runs != nil {
+			m.runs = msg.runs
+			m.readAt = time.Now()
+		}
+		if msg.err != nil {
+			m.err = msg.err
+		} else {
+			m.err = nil
+		}
+		return m, nil
+
+	case runLogMsg:
+		if msg.runID != "" {
+			m.logRunID = msg.runID
+		}
+		m.logText = msg.text
+		if msg.err != nil {
+			m.err = msg.err
+		}
+		return m, nil
+
+	case errMsg:
+		m.err = msg.err
+		return m, nil
+	}
+	return m, nil
+}
+
+// chosenRun is the run the roster cursor is on, in the runs tab.
+func (m agentsModel) chosenRun() *paperclip.AgentRun {
+	if m.cursor >= 0 && m.cursor < len(m.runs) {
+		return &m.runs[m.cursor]
+	}
+	return nil
+}
+
+// chosenAgent is the agent whose name matches the chosen run, or the first one.
+func (m agentsModel) chosenAgent() *paperclip.Agent {
+	if r := m.chosenRun(); r != nil {
+		for i := range m.agents {
+			if m.agents[i].ID == r.AgentID {
+				return &m.agents[i]
+			}
+		}
+	}
+	if len(m.agents) > 0 {
+		return &m.agents[0]
+	}
+	return nil
+}
+
+func (m agentsModel) View() string {
+	if m.width == 0 || m.height == 0 {
+		return "reading…"
+	}
+	if m.width < 80 || m.height < 24 {
+		return lipgloss.NewStyle().Width(m.width).Height(m.height).Render(
+			fmt.Sprintf("This workspace needs 80 × 24.\nThis panel is %d × %d.", m.width, m.height))
+	}
+
+	if m.showHelp {
+		return helpOverlay("agents", [][2]string{
+			{"[ ]  h l", "roster, runs, log"},
+			{"enter", "open the chosen run's log"},
+			{"p / P", "pause / resume the agent"},
+			{"w", "wake the agent now"},
+			{"↑ ↓  j k", "scroll"},
+			{"r / R", "read again / rebuild"},
+			{"q", "leave"},
+		}, m.width, m.height)
+	}
+
+	head := titleStyle.Render("Agents") + mutedStyle2().Render("  ·  who is working, on what, for how much")
+	state := "read " + m.readAt.Format("15:04:05")
+	if m.readAt.IsZero() {
+		state = "reading…"
+	}
+
+	var body []string
+	switch m.tab {
+	case 0:
+		body = m.rosterLines(m.width - 4)
+	case 1:
+		body = m.runLines(m.width - 4)
+	default:
+		body = m.logLines(m.width - 4)
+	}
+
+	p := panel{title: fmt.Sprintf("agents · %d agent(s), %d run(s)", len(m.agents), len(m.runs)),
+		width: m.width, height: m.height - 2, tabs: agentTabs, active: m.tab, focused: true}
+	visible := p.bodyHeight()
+	maxScroll := max(0, len(body)-visible)
+	if m.scroll > maxScroll {
+		m.scroll = maxScroll
+	}
+	if m.scroll < 0 {
+		m.scroll = 0
+	}
+	from := m.scroll
+	to := min(len(body), from+visible)
+
+	footer := hintLine([]string{
+		keyHint("[ ]", "roster, runs, log"),
+		keyHint("enter", "open a run"),
+		keyHint("p P w", "pause, resume, wake"),
+		keyHint("r", "read again"),
+		keyHint("?", "keys"),
+	}, m.width)
+	if m.notice != "" {
+		footer = lipgloss.NewStyle().Foreground(lipgloss.Color(active.Success)).Render(m.notice)
+	}
+	if m.err != nil {
+		footer = lipgloss.NewStyle().Foreground(lipgloss.Color(active.Error)).Render(truncateVisible("error: "+m.err.Error(), m.width))
+	}
+
+	gap := max(2, m.width-lipgloss.Width(head)-lipgloss.Width(state)-1)
+	return strings.Join([]string{
+		head + strings.Repeat(" ", gap) + mutedStyle2().Render(state),
+		p.render(body[from:to]),
+		footer,
+	}, "\n")
+}
+
+func (m agentsModel) rosterLines(width int) []string {
+	out := []string{"", dimStyle().Render(fmt.Sprintf(" %-16s %-10s %-18s %8s %8s", "agent", "state", "working on", "spent", "budget"))}
+	for _, a := range m.agents {
+		mark := icons.Todo
+		if a.Status == "idle" {
+			mark = icons.Done
+		}
+		working := "—"
+		for _, r := range m.runs {
+			if r.AgentID == a.ID && r.Status == "running" {
+				working = m.taskLabel(r)
+				mark = icons.Running
+				break
+			}
+		}
+		out = append(out, fmt.Sprintf(" %s %-16s %-10s %-18s %6d c %6d c",
+			mark, a.Name, a.Status, truncateVisible(working, 18), a.SpentCents, a.BudgetCents))
+	}
+	out = append(out, "")
+	out = append(out, dimStyle().Render(" c means cents. Budget is set in Paperclip; change it there, not here."))
+	return out
+}
+
+func (m agentsModel) taskLabel(r paperclip.AgentRun) string {
+	if r.Snapshot.IssueID != "" {
+		return short(r.Snapshot.IssueID)
+	}
+	if r.Snapshot.TaskKey != "" {
+		return short(r.Snapshot.TaskKey)
+	}
+	return "—"
+}
+
+func (m agentsModel) runLines(width int) []string {
+	if len(m.runs) == 0 {
+		return []string{"", mutedStyle2().Render("no runs recorded")}
+	}
+	out := []string{"", dimStyle().Render(fmt.Sprintf(" %-10s %-11s %-9s %-12s %-8s %s",
+		"when", "agent", "state", "woken by", "took", "task"))}
+	for _, r := range m.runs {
+		when := "-"
+		if t, err := time.Parse(time.RFC3339, r.StartedAt); err == nil {
+			when = t.Local().Format("15:04:05")
+		}
+		took := "-"
+		if r.FinishedAt != "" {
+			if s, err := time.Parse(time.RFC3339, r.StartedAt); err == nil {
+				if f, err := time.Parse(time.RFC3339, r.FinishedAt); err == nil {
+					took = f.Sub(s).Round(time.Second).String()
+				}
+			}
+		} else if r.Status == "running" {
+			took = "running"
+		}
+		colour := active.Text
+		switch r.Status {
+		case "running":
+			colour = active.Info
+		case "succeeded":
+			colour = active.Success
+		case "failed":
+			colour = active.Error
+		}
+		out = append(out, fmt.Sprintf(" %-10s %-11s %-9s %-12s %-8s %s",
+			when, truncateVisible(m.names[r.AgentID], 11),
+			lipgloss.NewStyle().Foreground(lipgloss.Color(colour)).Render(fmt.Sprintf("%-9s", r.Status)),
+			r.Source, took, m.taskLabel(r)))
+	}
+	return out
+}
+
+// logEntry is one readable line out of a run log.
+type logEntry struct {
+	when   string
+	stream string
+	line   string
+}
+
+// decodeRunLog turns what Paperclip stores into something a person can read.
+//
+// A run log is not lines of prose: it is an envelope around newline-separated
+// JSON, and most of those are token-by-token streaming events. Printed raw they
+// are unreadable and they redraw the panel. So deltas are gathered back into
+// paragraphs, and everything else becomes one labelled line.
+func decodeRunLog(raw string) []logEntry {
+	var envelope struct {
+		Content string `json:"content"`
+	}
+	content := raw
+	if json.Unmarshal([]byte(raw), &envelope) == nil && envelope.Content != "" {
+		content = envelope.Content
+	}
+
+	var out []logEntry
+	var buf strings.Builder
+	bufWhen := ""
+	flush := func() {
+		text := strings.TrimSpace(cleanText(buf.String()))
+		buf.Reset()
+		if text == "" {
+			return
+		}
+		for _, para := range strings.Split(text, "\n") {
+			para = strings.TrimSpace(para)
+			if para != "" {
+				out = append(out, logEntry{when: bufWhen, stream: "thinking", line: para})
+			}
+		}
+	}
+
+	for _, line := range strings.Split(content, "\n") {
+		line = strings.TrimSpace(line)
+		if line == "" {
+			continue
+		}
+		var rec struct {
+			TS     string `json:"ts"`
+			Stream string `json:"stream"`
+			Chunk  string `json:"chunk"`
+		}
+		text := line
+		when := ""
+		if json.Unmarshal([]byte(line), &rec) == nil && rec.Chunk != "" {
+			text, when = rec.Chunk, shortTime(rec.TS)
+		}
+
+		var ev struct {
+			Type    string `json:"type"`
+			Delta   string `json:"delta"`
+			Message struct {
+				Role    string `json:"role"`
+				Content []struct {
+					Type string `json:"type"`
+					Text string `json:"text"`
+				} `json:"content"`
+			} `json:"message"`
+			AssistantMessageEvent struct {
+				Type  string `json:"type"`
+				Delta string `json:"delta"`
+			} `json:"assistantMessageEvent"`
+		}
+
+		if json.Unmarshal([]byte(text), &ev) != nil {
+			// Not JSON we know: show it as a line, trimmed.
+			if t := strings.TrimSpace(cleanText(text)); t != "" {
+				out = append(out, logEntry{when: when, line: t})
+			}
+			continue
+		}
+
+		kind := ev.AssistantMessageEvent.Type
+		if ev.Type == "message_update" && strings.HasSuffix(kind, "_delta") {
+			if bufWhen == "" {
+				bufWhen = when
+			}
+			buf.WriteString(ev.AssistantMessageEvent.Delta)
+			continue
+		}
+
+		// Anything else ends the paragraph that was being gathered.
+		flush()
+		bufWhen = when
+
+		switch ev.Type {
+		case "message_start", "message_end":
+			role := ev.Message.Role
+			if role == "" {
+				continue
+			}
+			for _, c := range ev.Message.Content {
+				t := strings.TrimSpace(cleanText(c.Text))
+				if t == "" {
+					continue
+				}
+				if i := strings.Index(t, "\n"); i > 0 {
+					t = t[:i]
+				}
+				out = append(out, logEntry{when: when, stream: "prompt", line: role + ": " + t})
+				break
+			}
+		case "tool_call", "tool_result", "tool_use":
+			out = append(out, logEntry{when: when, stream: "tool", line: ev.Type})
+		case "turn_start", "turn_end", "session", "agent_start", "agent_end":
+			out = append(out, logEntry{when: when, stream: "event", line: ev.Type})
+		default:
+			if kind != "" {
+				out = append(out, logEntry{when: when, stream: "event", line: kind})
+			}
+		}
+	}
+	flush()
+
+	if len(out) == 0 {
+		for _, line := range strings.Split(strings.TrimRight(content, "\n"), "\n") {
+			if t := strings.TrimSpace(cleanText(line)); t != "" {
+				out = append(out, logEntry{line: t})
+			}
+		}
+	}
+	return out
+}
+
+// shortTime is the clock time of a log line, in the reader's timezone.
+func shortTime(ts string) string {
+	if t, err := time.Parse(time.RFC3339Nano, ts); err == nil {
+		return t.Local().Format("15:04:05")
+	}
+	return ""
+}
+
+// cleanText removes escape sequences and control characters, so a log cannot
+// repaint the panel it is drawn in.
+func cleanText(s string) string {
+	var b strings.Builder
+	skip := false
+	for _, r := range s {
+		if r == 0x1b {
+			skip = true
+			continue
+		}
+		if skip {
+			if (r >= 'a' && r <= 'z') || (r >= 'A' && r <= 'Z') {
+				skip = false
+			}
+			continue
+		}
+		if r == '\t' {
+			b.WriteString("  ")
+			continue
+		}
+		if r < 0x20 && r != '\n' {
+			continue
+		}
+		b.WriteRune(r)
+	}
+	return b.String()
+}
+
+func (m agentsModel) logLines(width int) []string {
+	if m.logRunID == "" {
+		return []string{"", mutedStyle2().Render("choose a run on the runs tab, then press enter")}
+	}
+	if m.logText == "" {
+		return []string{"", mutedStyle2().Render("no output recorded for this run")}
+	}
+	var out []string
+	out = append(out, "", dimStyle().Render(" run "+short(m.logRunID)))
+	entries := decodeRunLog(m.logText)
+	if len(entries) > 2000 {
+		entries = entries[len(entries)-2000:]
+	}
+	for _, e := range entries {
+		if lipgloss.Width(e.line) > width {
+			e.line = truncateVisible(e.line, width)
+		}
+		colour := active.Text
+		switch e.stream {
+		case "thinking":
+			colour = active.Muted
+		case "prompt":
+			colour = active.Info
+		case "tool":
+			colour = active.Warning
+		case "event":
+			colour = active.Overlay
+		}
+		out = append(out, " "+dimStyle().Render(e.when)+" "+
+			lipgloss.NewStyle().Foreground(lipgloss.Color(colour)).Render(e.line))
+	}
+	return out
+}

tui/frame.go

@@ -289,3 +289,12 @@ func helpOverlay(title string, entries [][2]string, w, h int) string {
 	p := panel{title: title, width: boxW, height: boxH, focused: true}
 	return lipgloss.Place(w, h, lipgloss.Center, lipgloss.Center, p.render(lines))
 }
+
+// short is the first eight characters of an identifier: enough to recognise a
+// run or a task in a column, never enough to confuse two of them on one screen.
+func short(id string) string {
+	if len(id) <= 8 {
+		return id
+	}
+	return id[:8]
+}

tui/main.go

@@ -19,7 +19,7 @@ import (
 const defaultAgent = "403de1f7-645c-4d35-bc7e-19b01d77bacc"
 
 func main() {
-	screen := flag.String("screen", "board", "which screen to run: board, work, reply, new, project, repair")
+	screen := flag.String("screen", "board", "which screen to run: board, work, reply, new, agents, project, repair")
 	folder := flag.String("folder", "", "project folder, for the board screen (defaults to the current directory)")
 	light := flag.String("theme", "", "colour variant: dark, light, or empty to follow the terminal")
 	ascii := flag.Bool("ascii", false, "plain characters instead of Nerd Font icons")
@@ -54,6 +54,13 @@ func main() {
 		}
 		m = newBoardModel(client, proj, dir)
 
+	case "agents":
+		dir := *folder
+		if dir == "" {
+			dir, _ = os.Getwd()
+		}
+		m = newAgentsModel(client, dir)
+
 	case "project":
 		dir := *folder
 		if dir == "" {
@@ -127,7 +134,7 @@ func main() {
 		m = newReplyModel(client, dir, id)
 
 	default:
-		fmt.Fprintf(os.Stderr, "unknown screen %q: use board, work, reply, new, project or repair\n", *screen)
+		fmt.Fprintf(os.Stderr, "unknown screen %q: use board, work, reply, new, agents, project or repair\n", *screen)
 		os.Exit(1)
 	}
 

tui/paperclip/client.go

@@ -373,3 +373,71 @@ func (c *Client) CostSummary(issueID string) (Cost, error) {
 func (c *Client) Health() error {
 	return c.do("GET", "/api/health", nil, nil)
 }
+
+// AgentRun is one heartbeat run in the company, whichever agent ran it.
+type AgentRun struct {
+	ID              string `json:"id"`
+	AgentID         string `json:"agentId"`
+	Status          string `json:"status"`
+	Source          string `json:"invocationSource"`
+	StartedAt       string `json:"createdAt"`
+	FinishedAt      string `json:"finishedAt"`
+	ErrorCode       string `json:"errorCode"`
+	LastOutputAt    string `json:"lastOutputAt"`
+	LastOutputBytes int    `json:"lastOutputBytes"`
+	// The task a run worked on lives inside the run's context snapshot, not on
+	// the run itself.
+	Snapshot struct {
+		IssueID string `json:"issueId"`
+		TaskKey string `json:"taskKey"`
+	} `json:"contextSnapshot"`
+}
+
+// Runs returns every run in the company, newest first.
+func (c *Client) CompanyRuns() ([]AgentRun, error) {
+	var out []AgentRun
+	err := c.do("GET", "/api/companies/"+c.Company+"/heartbeat-runs", nil, &out)
+	return out, err
+}
+
+// RunLog returns the recorded output of one run.
+func (c *Client) RunLog(runID string, maxBytes int) (string, error) {
+	var raw []byte
+	req, err := http.NewRequest("GET", c.BaseURL+"/api/heartbeat-runs/"+runID+"/log", nil)
+	if err != nil {
+		return "", err
+	}
+	req.Header.Set("Authorization", "Bearer "+c.token)
+	resp, err := c.http.Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode < 200 || resp.StatusCode > 299 {
+		b, _ := io.ReadAll(io.LimitReader(resp.Body, 400))
+		return "", fmt.Errorf("run log returned %d: %s", resp.StatusCode, snippet(b))
+	}
+	raw, err = io.ReadAll(io.LimitReader(resp.Body, int64(maxBytes)))
+	if err != nil {
+		return "", err
+	}
+	if len(raw) == maxBytes {
+		return string(raw) + "\n… (only the last " + fmt.Sprint(maxBytes) + " bytes are shown)", nil
+	}
+	return string(raw), nil
+}
+
+// PauseAgent and ResumeAgent are the two ways to stop and start an agent.
+func (c *Client) PauseAgent(agentID string) error {
+	return c.do("POST", "/api/agents/"+agentID+"/pause", map[string]any{}, nil)
+}
+
+func (c *Client) ResumeAgent(agentID string) error {
+	return c.do("POST", "/api/agents/"+agentID+"/resume", map[string]any{}, nil)
+}
+
+// WakeAgent asks Paperclip to start a run for this agent now.
+func (c *Client) WakeAgent(agentID string) error {
+	return c.do("POST", "/api/agents/"+agentID+"/wakeup",
+		map[string]any{"source": "on_demand", "triggerDetail": "manual", "forceFreshSession": false}, nil)
+}