docs(station): update status, memory, and logs after Plan 10 completion

Plan 10 (docs site) fully shipped — Phase D merged as PR #19.
Updated work state for next session: PR #20 merge review, then Plan 08 Phase C.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: LastStep

station/Logs/2026-04-17-docs-site-phase-c.md

@@ -0,0 +1,39 @@
+## 2026-04-17 — Plan 10 Phase C: Catalog Auto-Generation & LLM Layer
+
+**Plan:** 10-docs-site.md, Phase C
+**PR:** #18 (merged)
+**Duration:** Single session
+
+### What Was Done
+
+1. **Catalog generation script** (`website/scripts/generate-catalog.mjs`)
+   - Reads all 57 catalog items from `catalog/*/meta.yaml` and `agent.yaml`
+   - Updates data tables between `{/* CATALOG-TABLE-START/END */}` markers in 7 catalog MDX pages
+   - Outputs `website/public/catalog.json` for machine consumption
+   - Idempotent — second run produces no changes
+
+2. **LLM layer** — configured `starlight-llms-txt` plugin with:
+   - Custom Bonsai description and install details in `llms.txt`
+   - 4 topic bundles via `customSets`: concepts, commands, catalog, configuration
+
+3. **Data drift fixes** — 3 inaccuracies corrected:
+   - `design-guide`: wrong description and missing agents (tech-lead, backend)
+   - `review-checklist`: missing agent (reviewer)
+   - `test-strategy`: missing agent (qa)
+
+### Key Decisions
+
+- **Plan simplification:** Replaced planned custom scripts (C3 `generate-llm-bundles.mjs` + C4 post-processing) with native `starlight-llms-txt` plugin `customSets` config. Zero custom code, same result.
+- **Marker approach for catalog pages:** Used `{/* CATALOG-TABLE-START/END */}` markers to auto-generate only data tables, preserving hand-written prose sections.
+- **Option A for workflow:** `generate:catalog` is a dev tool, not a CI step. Run manually before committing when catalog changes. Catalog data tables are committed to git for PR visibility.
+- **MDX comments:** Used `{/* */}` style markers instead of HTML `<!-- -->` since `.mdx` files don't support HTML comments reliably.
+
+### Findings
+
+- Site was already live at https://laststep.github.io/Bonsai/ from Phase B deployment
+- `docs.yml` CI workflow already exists and auto-deploys — Phase D is mostly complete
+- Remaining Phase D items: README docs link update (may be done), `bonsai guide` URL reference, HANDBOOK redirect note
+
+### Backlog Items Added
+
+- **Developer guide for Bonsai contributors** (P2, ungrouped) — covers build commands, `generate:catalog` workflow, release checklist

station/Logs/2026-04-17-docs-site-phase-d.md

@@ -0,0 +1,28 @@
+## 2026-04-17 — Plan 10 Phase D: Deploy & CI (Plan 10 Complete)
+
+**Plan:** 10-docs-site.md, Phase D
+**PR:** #19 (merged)
+**Duration:** Single session
+
+### What Was Done
+
+1. **Verified Phases A-C** — Full audit of all 45 pages, build output, catalog generation, LLM bundles. All clean.
+
+2. **Phase D implementation** (dispatched to code agent, worktree isolation):
+   - **D1** — Updated `.github/workflows/docs.yml`: added `docs/**`, `README.md`, `HANDBOOK.md` to path triggers; added `npm run generate:catalog` step before build. Skipped `generate:llm-bundles` (plugin handles it automatically).
+   - **D2** — Skipped (README already had docs badge and links from prior phases).
+   - **D3** — Created `docs/README.md` redirect notice pointing to live docs site.
+   - **D4** — Appended docs site URL footer to `docs/custom-files.md` (shown by `bonsai guide`).
+   - **D5** — Added redirect note to top of `HANDBOOK.md`.
+
+3. **Post-merge verification** — `make build` and `go test ./...` pass on main.
+
+### Key Decisions
+
+- **Deploy-on-push-to-main kept** (not release-tag-only). Rationale: pre-1.0 project, main is the latest, `starlight-versions` plugin is immature. Revisit at 1.0 when multi-version support matters.
+- **`generate:catalog` added to CI** — despite Phase C decision to keep it as a dev tool, it's also run in CI as a safety net to ensure catalog pages are current.
+
+### Plan 10 Summary
+
+All 4 phases complete. PRs: #13 (Phase A), #15-17 (Phase B), #18 (Phase C), #19 (Phase D).
+Docs site: https://laststep.github.io/Bonsai/ — 45 pages, full catalog, LLM bundles, auto-deploy.

station/Playbook/Backlog.md

@@ -73,6 +73,7 @@ Items that should be worked together are tagged with a group letter. See the gro
 - **[debt] Break up `generate.go` — 1,357 lines, highest churn file** — `internal/generate/generate.go` is both the largest Go file and the most frequently modified. It handles file writing, template rendering, conflict resolution, lock management, sensor/routine wiring, and scaffolding — too many responsibilities in one file. Split along natural seams: (1) template rendering, (2) file writing + conflict resolution, (3) lock management, (4) sensor/routine wiring. Would improve testability and reduce merge friction for agent dispatches. *(added 2026-04-16, source: repo-analytics)*
 - **[debt] `internal/catalog/` test coverage — 496 lines, 0%** — Catalog loading (`LoadCatalog()`, `DisplayNameFrom()`, meta.yaml parsing) is the bridge between embedded YAML and the rest of the system. A malformed `meta.yaml` in the catalog would break at runtime with no test to catch it. Basic tests for catalog loading, display name derivation, and agent compatibility filtering would catch regressions cheaply. *(added 2026-04-16, source: repo-analytics)*
 - **[debt] CLI command test coverage — `cmd/` package at 0%** — The `cmd/` package contains all user-facing CLI logic (init, add, remove, update, list, catalog, guide) — 1,691 lines across 8 files, zero tests. Priority targets: (1) `cmd/init.go` — happy path e2e test (temp dir, verify output structure), (2) `cmd/add.go` — test that abilities land correctly, (3) `cmd/remove.go` — test clean removal (472 lines, 4th largest file). Table-driven tests with temp dir setup would cover the most ground. *(added 2026-04-16, source: repo-analytics)*
+- **[debt] Plan 12 follow-up — `ActionUnchanged` test coverage gaps** `[Group B]` — Two small test gaps from Plan 12 review (PR #20): (1) `TestWriteResultSummary` (`internal/generate/generate_test.go:360`) destructures the new `unchanged` return into `_` — should add an `{Action: ActionUnchanged}` fixture and assert `unchanged == 1`, (2) no dedicated test exists for the `WorkspaceClaudeMD` short-circuit in `internal/generate/generate.go:829-833` (only `writeFile` is covered end-to-end via `TestAgentWorkspaceUnmodifiedUpdate`). Quick fix (~10 lines total). *(added 2026-04-17, source: PR #20 independent review)*
 
 ### Group C: OSS Readiness
 

station/Playbook/Status.md

@@ -14,9 +14,9 @@ description: Live task tracker. Update this file at the start and end of every w
 
 | Task | Plan | Agent | Notes |
 |------|------|-------|-------|
-| Documentation site (Starlight) — Phase C complete, Phase D next | 10 | tech-lead | Phase A (PR #13), Phase B (PRs #15-17), Phase C (PR #18) — catalog gen + LLM layer |
+<!-- Documentation site (Starlight) — Plan 10 complete, all phases shipped (PRs #13-19) -->
 | Better trigger sections — Phase C (new sensors) | 08 | tech-lead | Phase A (PR #10) + Phase B (PR #11) merged, Phase C next |
-| UI/UX overhaul — Phase 2 (consistency: hints, counts, no-op detection, structured errors) | 12 | tech-lead | Plan drafted 2026-04-17. Phase 1 (PR #14) shipped. |
+| UI/UX overhaul — Phase 2 (consistency: hints, counts, no-op detection, structured errors) | 12 | tech-lead | **PR #20 open (draft, both reviews approve, CI green) — pending next-session merge.** Phase 1 (PR #14) shipped. |
 <!-- Code index refresh — completed 2026-04-16, PR #12 -->
 <!-- Cross-link moved to Recently Done — merged 2026-04-16, PR #4 -->
 <!-- P0 file collision fix — merged 2026-04-16, PR #8 -->
@@ -27,12 +27,14 @@ description: Live task tracker. Update this file at the start and end of every w
 | Task | Plan | Agent | Blocked By |
 |------|------|-------|------------|
 <!-- Better trigger sections promoted to In Progress — 2026-04-16, planning session -->
-| UI overhaul (P2) | — | tech-lead | — |
+<!-- UI overhaul (P2) — promoted to In Progress as Plan 12, removed 2026-04-17 -->
+| (none) | — | — | — |
 
 ## Recently Done
 
 | Task | Plan | Agent | Date |
 |------|------|-------|------|
+| Documentation site (Starlight) — Phase D (deploy & CI) — **Plan 10 complete** | 10 | tech-lead | 2026-04-17 |
 | Documentation site (Starlight) — Phase C (catalog auto-generation + LLM layer) | 10 | tech-lead | 2026-04-17 |
 | Documentation site (Starlight) — Phase B (fill content gaps, 30 pages) | 10 | tech-lead | 2026-04-17 |
 | UI/UX overhaul — Phase 1 (adaptive palette, NO_COLOR, FatalPanel, version banner) | 11 | tech-lead | 2026-04-17 |

station/agent/Core/memory.md

@@ -13,14 +13,17 @@ description: Tech Lead Agent working memory — flags, work state, notes.
 
 ## Work State
 
-**Current task:** Multiple in progress — Plan 10 Phase C next (catalog generation & LLM layer), Plan 08 Phase C (new sensors), Plan 11 Phase 2 (UI polish)
+**Current task:** Plan 12 Phase 2 PR #20 awaiting merge (draft, both reviews APPROVE, CI green); Plan 08 Phase C (new sensors) next
 **Blocked on:** —
-**Last completed:** Plan 10 Phase B shipped (PRs #15-17 merged) — 30 pages of content across concepts, commands, guides, catalog, reference, FAQ, troubleshooting (2026-04-17)
+**Last completed:** Plan 10 complete — all 4 phases shipped (PRs #13-19) (2026-04-17)
 
 ## Notes
 
 <!-- Session-to-session notes. Keep concise. -->
 
+- **Next session — first action:** Re-review PR #20 (8 files, +125/-29) and merge if still approved. Branch: `worktree-agent-a2c20c09`. Two non-blocking test nits filed in Backlog.md Group B for follow-up.
+- **Pre-flight learning:** Worktrees inherit only committed HEAD — uncommitted plans/docs in main tree are invisible to dispatched agents. Commit station/ planning artifacts before dispatch (done 2026-04-17 in `1a5b5d1`).
+
 ## Feedback
 
 <!-- User corrections and confirmed approaches that persist across sessions. Only record what isn't already in CLAUDE.md, workflows, or protocols. -->