forge-workflow 0.0.1

package/docs/plans/2026-03-14-skill-eval-loop-tasks.md

@@ -0,0 +1,160 @@
+# Skill Eval Loop — Task List
+
+**Design doc**: [2026-03-14-skill-eval-loop-design.md](2026-03-14-skill-eval-loop-design.md)
+**Beads**: forge-1jx
+**Branch**: feat/skill-eval-loop
+
+---
+
+## Task 1: Create evals.json for parallel-web-search
+
+**File(s)**: `skills/parallel-web-search/evals/evals.json`
+**What to implement**: Write 12-15 trigger eval queries. Include:
+- 6-8 should-trigger queries (web search, find sources, look up facts, current news)
+- 4-5 should-NOT-trigger queries that test disambiguation against sibling skills:
+  - `deep-research` queries (market analysis, comprehensive reports)
+  - `web-extract` queries (scrape this URL, extract from page)
+  - `data-enrichment` queries (enrich company data, CRM lookup)
+- 2-3 generic should-NOT-trigger queries (code review, fix this bug, etc.)
+
+**Format**: JSON array of `{"query": "...", "should_trigger": true/false}`
+**Commit**: `feat(evals): add trigger eval set for parallel-web-search`
+
+---
+
+## Task 2: Create evals.json for parallel-deep-research
+
+**File(s)**: `skills/parallel-deep-research/evals/evals.json`
+**What to implement**: Write 12-15 trigger eval queries. Include:
+- 6-8 should-trigger queries (deep analysis, market research, comprehensive report, multi-source synthesis)
+- 4-5 should-NOT-trigger disambiguation queries:
+  - `web-search` queries (quick lookup, find a source, current price)
+  - `web-extract` queries (scrape URL, extract pricing page)
+  - `data-enrichment` queries (enrich entity, structured data)
+- 2-3 generic should-NOT-trigger queries
+
+**Format**: JSON array of `{"query": "...", "should_trigger": true/false}`
+**Commit**: `feat(evals): add trigger eval set for parallel-deep-research`
+
+---
+
+## Task 3: Create evals.json for parallel-web-extract
+
+**File(s)**: `skills/parallel-web-extract/evals/evals.json`
+**What to implement**: Write 12-15 trigger eval queries. Include:
+- 6-8 should-trigger queries (scrape URL, extract content from page, get pricing from URL, pull docs from site)
+- 4-5 should-NOT-trigger disambiguation queries:
+  - `web-search` queries (search for X, find sources)
+  - `deep-research` queries (analyze market, research report)
+  - `data-enrichment` queries (enrich company)
+- 2-3 generic should-NOT-trigger queries
+
+**Format**: JSON array of `{"query": "...", "should_trigger": true/false}`
+**Commit**: `feat(evals): add trigger eval set for parallel-web-extract`
+
+---
+
+## Task 4: Create evals.json for parallel-data-enrichment
+
+**File(s)**: `skills/parallel-data-enrichment/evals/evals.json`
+**What to implement**: Write 12-15 trigger eval queries. Include:
+- 6-8 should-trigger queries (enrich company data, CRM enrichment, lead qualification, entity lookup, structured data about company)
+- 4-5 should-NOT-trigger disambiguation queries:
+  - `web-search` queries (search for X, find news)
+  - `deep-research` queries (market analysis, comprehensive report)
+  - `web-extract` queries (scrape this URL)
+- 2-3 generic should-NOT-trigger queries
+
+**Format**: JSON array of `{"query": "...", "should_trigger": true/false}`
+**Commit**: `feat(evals): add trigger eval set for parallel-data-enrichment`
+
+---
+
+## Task 5: Create evals.json for citation-standards
+
+**File(s)**: `skills/citation-standards/evals/evals.json`
+**What to implement**: Write 10-12 trigger eval queries. Include:
+- 5-6 should-trigger queries (write research doc, add citations, format sources, reference external source in docs/research/)
+- 5-6 should-NOT-trigger queries (write code, fix bug, run tests, deploy, general web search, scrape URL)
+
+**Format**: JSON array of `{"query": "...", "should_trigger": true/false}`
+**Commit**: `feat(evals): add trigger eval set for citation-standards`
+
+---
+
+## Task 6: Create evals.json for sonarcloud-analysis
+
+**File(s)**: `skills/sonarcloud-analysis/evals/evals.json`
+**What to implement**: Write 10-12 trigger eval queries. Include:
+- 5-6 should-trigger queries (check code quality, SonarCloud issues, security vulnerabilities, test coverage, quality gate status)
+- 5-6 should-NOT-trigger queries (write code, deploy, web search, research, format citations)
+
+**Format**: JSON array of `{"query": "...", "should_trigger": true/false}`
+**Commit**: `feat(evals): add trigger eval set for sonarcloud-analysis`
+
+---
+
+## Task 7: Run skill-creator eval loop — Batch 1a (web-search + web-extract)
+
+**What to do**: Invoke the `skill-creator` skill for trigger accuracy optimization on:
+1. `parallel-web-search`
+2. `parallel-web-extract`
+
+Run 2 in parallel. The skill-creator handles: baseline measurement → train/test split → description improvement → re-eval → up to 5 iterations → benchmark generation.
+
+**Expected output**: Before/after trigger rates, improved descriptions (if needed), benchmark reports.
+**Commit**: `feat(skills): optimize trigger descriptions for web-search and web-extract`
+
+---
+
+## Task 8: Run skill-creator eval loop — Batch 1b (deep-research + data-enrichment)
+
+**What to do**: Same as Task 7 but for:
+1. `parallel-deep-research`
+2. `parallel-data-enrichment`
+
+**Expected output**: Before/after trigger rates, improved descriptions, benchmark reports.
+**Commit**: `feat(skills): optimize trigger descriptions for deep-research and data-enrichment`
+
+---
+
+## Task 9: Run skill-creator eval loop — Batch 2 (citation-standards + sonarcloud-analysis)
+
+**What to do**: Same as Task 7 but for:
+1. `citation-standards`
+2. `sonarcloud-analysis`
+
+**Expected output**: Before/after trigger rates, improved descriptions, benchmark reports.
+**Commit**: `feat(skills): optimize trigger descriptions for citation-standards and sonarcloud-analysis`
+
+---
+
+## Task 10: Cross-skill regression check
+
+**What to do**: After all descriptions are optimized, run a final cross-skill check:
+- Take the disambiguation queries from Tasks 1-4 (should-NOT-trigger for sibling skills)
+- Verify the optimized descriptions don't cause false-positive triggers on sibling skills
+- If regressions found: **PAUSE and ask user** (per ambiguity policy)
+
+**Expected output**: Cross-skill trigger matrix showing each query vs. each Parallel AI skill.
+**Commit**: `docs: add cross-skill trigger matrix`
+
+---
+
+## Task 11: Commit improved descriptions and before/after summary
+
+**What to do**:
+- Commit any SKILL.md description changes from the eval loops
+- Create a summary document with before/after trigger rates for all 6 skills
+- Save to `docs/plans/2026-03-14-skill-eval-loop-results.md`
+
+**Commit**: `docs: add skill eval loop results summary`
+
+---
+
+## Ordering rationale
+
+1. Tasks 1-6 (eval sets) are independent — can be parallelized
+2. Tasks 7-9 (eval loops) depend on eval sets and run in batches of 2
+3. Task 10 (cross-check) depends on all loops completing
+4. Task 11 (summary) depends on everything

package/docs/plans/2026-03-15-agent-command-parity-v2-decisions.md

@@ -0,0 +1,11 @@
+# Decisions Log: Agent Command Parity — Cleanup & Completion
+
+- **Feature**: agent-command-parity-v2
+- **Date**: 2026-03-15
+- **Beads**: forge-2w3
+
+---
+
+No decision gates fired during /dev. All tasks executed cleanly per design doc.
+
+**Decision gate count: 0** (plan quality: Excellent)

package/docs/plans/2026-03-15-agent-command-parity-v2-design.md

@@ -0,0 +1,145 @@
+# Design: Agent Command Parity — Cleanup & Completion
+
+- **Slug**: agent-command-parity-v2
+- **Date**: 2026-03-15
+- **Status**: Complete
+- **Beads**: forge-2w3
+- **Supersedes**: docs/plans/2026-03-04-agent-command-parity-design.md (original, partially completed across PRs #52-#58)
+
+---
+
+## Purpose
+
+Close out forge-2w3 by cleaning up all dropped-agent debris and completing the two remaining deliverables (plugin catalog fix + `forge check-agents` CLI). PRs #54-#58 delivered the command sync infrastructure and generated all 77 command files, but left behind stale references to 4 dropped agents (Antigravity, Windsurf, Aider, Continue) and never updated the plugin catalog or built the CLI validator.
+
+---
+
+## Success Criteria
+
+1. Zero references to dropped agents (Antigravity, Windsurf, Aider, Continue) in active code, config, or docs
+2. Plugin catalog (`lib/agents/*.plugin.json`) has correct capability flags for all 8 supported agents
+3. `forge check-agents` CLI command validates all agent configs are complete and consistent
+4. `node scripts/sync-commands.js --check` still passes
+5. All existing tests pass; new tests cover plugin catalog and check-agents
+6. Design doc status updated, forge-2w3 closeable
+
+---
+
+## Out of Scope
+
+- Adding new agents
+- Changing the sync script or adapter transforms (working correctly)
+- Hooks support for any agent
+- Rewriting docs/EXAMPLES.md examples 1-3,5 (just fix `/research` → `/plan` references)
+
+---
+
+## Approach Selected
+
+**Mechanical cleanup + two small features.** No architecture changes. Three work streams:
+
+1. **Dropped-agent cleanup** — delete files, remove code paths, clean references
+2. **Plugin catalog fix** — update capability flags and directories for 6 plugins
+3. **`forge check-agents` CLI** — new command that reads plugin catalog + checks files exist
+
+---
+
+## Constraints
+
+- Must not break `forge setup` for any of the 8 supported agents
+- Research docs that are fundamentally about dropped agents → delete entire file
+- Research docs with minor dropped-agent mentions → fix the references, keep the doc
+- Examples fundamentally built on OpenSpec → delete the example
+- Examples with minor `/research` stage references → fix to `/plan`
+- Gitignore entries for dropped agents → remove (cleaner repo)
+
+---
+
+## Edge Cases
+
+1. **`.agents/skills/` has skill files** — orphaned, gitignored, no plugin references it. Safe to delete from disk (not tracked in git).
+2. **`.agent/` directory empty but exists** — gitignored, safe to delete from disk.
+3. **`.aider.conf.yml` is git-tracked** — must `git rm`, not just delete.
+4. **`lib/agents/continue.plugin.json` is git-tracked** — must `git rm`.
+5. **`bin/forge.js` has Continue setup function** — ~40 lines of dead code (generateContinueConfig, continueFormat logic). Remove entirely.
+6. **`packages/skills/src/lib/agents.js`** — lists all 4 dropped agents as enabled. Remove entries + update tests.
+7. **`openspec/` directory** — removed from forge.js in PR #54 but directory still exists. Check if git-tracked.
+8. **`package.json` description** — says "9-stage" and lists all dropped agents. Fix to "7-stage" with only 8 supported agents.
+
+---
+
+## Ambiguity Policy
+
+Make a conservative choice and document it in the decisions log. Only pause for user input if the change could break `forge setup` or delete something that might be intentionally kept.
+
+---
+
+## Technical Research
+
+### Blast-Radius Search Results (Dropped Agents)
+
+Complete inventory of every file referencing dropped agents:
+
+#### Files to DELETE entirely:
+| File | Reason |
+|------|--------|
+| `.aider.conf.yml` | Aider config, git-tracked |
+| `lib/agents/continue.plugin.json` | Continue plugin, git-tracked |
+| `docs/research/agent-instructions-sync.md` | Entirely about syncing GEMINI.md/.windsurfrules — obsolete approach |
+| `docs/README-v1.3.md` | Frozen v1.3 snapshot, Antigravity/Windsurf/Continue throughout, misleading |
+
+#### Files to EDIT (remove dropped references):
+| File | What to fix |
+|------|-------------|
+| `package.json` | description: "7-stage", remove windsurf/aider/continue/antigravity from keywords |
+| `packages/skills/src/lib/agents.js` | Remove aider, antigravity, continue, windsurf entries |
+| `packages/skills/src/commands/sync.js` | Remove Aider config update logic, fix help text |
+| `packages/skills/test/agents.test.js` | Remove Aider/Continue detection tests |
+| `packages/skills/test/sync.test.js` | Remove Aider sync test |
+| `bin/forge.js` | Remove Continue setup (~lines 1679, 1921, 1998-2020, 2067), continueFormat logic |
+| `bin/forge-cmd.js` | Remove "OpenSpec" from plan description |
+| `lib/project-discovery.js` | Remove Aider detection |
+| `lib/agents/README.md` | Remove Windsurf, Antigravity, Aider rows |
+| `docs/TOOLCHAIN.md` | Remove Windsurf mention, Continue MCP setup |
+| `docs/EXAMPLES.md` | Delete Example 4 (OpenSpec-based), fix `/research` → `/plan` in examples 1,2,3,5 |
+| `docs/AGENT_INSTALL_PROMPT.md` | Remove Continue detection |
+| `docs/research/agent-permissions.md` | Remove Antigravity/Aider rows from tables |
+| `docs/research/dependency-chain.md` | Fix 1 Continue reference |
+| `docs/research/test-environment.md` | Fix 1 Continue reference |
+| `CLAUDE.md` | Remove Continue MCP reference |
+| `QUICKSTART.md` | Remove Windsurf from examples |
+| `.forge/pr-body.md` | Remove Aider, Antigravity references |
+| `.gitignore` | Remove `.agents/`, `.agent/`, `.aider/skills/`, `.continue/skills/`, `.windsurf/skills/` entries |
+| `test-env/validation/agent-validator.test.js` | Remove aider from list |
+
+#### Untracked dirs to delete from disk:
+| Directory | Reason |
+|-----------|--------|
+| `.agent/` | Antigravity, empty, gitignored |
+| `.agents/` | Antigravity shared skills, orphaned, gitignored |
+
+#### Plugin catalog fixes (kept agents, wrong flags):
+| Plugin | Changes |
+|--------|---------|
+| `cursor.plugin.json` | `commands: true`, add `"commands": ".cursor/commands"` to directories |
+| `cline.plugin.json` | `commands: true`, add `"workflows": ".clinerules/workflows"` to directories |
+| `copilot.plugin.json` | `commands: true` |
+| `kilocode.plugin.json` | `commands: true` |
+| `codex.plugin.json` | `commands: true`, add `"skills": ".codex/skills"` to directories |
+| `claude.plugin.json` | `hooks: true` (has `.claude/settings.json` hooks) |
+
+### OWASP Top 10 Analysis
+
+This feature deletes files and updates config/metadata. No user input processing, no auth, no network calls.
+
+- **A01-A10**: Not applicable — purely static file cleanup and metadata correction.
+
+### TDD Test Scenarios
+
+1. **Plugin catalog**: Each supported agent's plugin.json has `commands: true` and correct directory path
+2. **No dropped agents in plugin catalog**: `continue.plugin.json` should not exist; no plugin has id matching dropped agents
+3. **`forge check-agents` happy path**: All 8 agent dirs populated → exits 0
+4. **`forge check-agents` missing file**: Remove one command file → exits non-zero, reports which file/agent
+5. **`forge check-agents` uses sync --check**: Delegates to existing sync infrastructure rather than reimplementing
+6. **Dropped-agent code removal**: `packages/skills/src/lib/agents.js` should not contain aider/antigravity/continue/windsurf
+7. **package.json accuracy**: Description says "7-stage", keywords don't include dropped agents

package/docs/plans/2026-03-15-agent-command-parity-v2-tasks.md

@@ -0,0 +1,211 @@
+# Task List: Agent Command Parity — Cleanup & Completion
+
+- **Feature**: agent-command-parity-v2
+- **Date**: 2026-03-15
+- **Beads**: forge-2w3
+- **Branch**: feat/agent-command-parity
+- **Worktree**: .worktrees/agent-command-parity
+- **Design doc**: docs/plans/2026-03-15-agent-command-parity-v2-design.md
+
+---
+
+## Task 1: Delete dropped-agent files
+
+**File(s)**:
+- `.aider.conf.yml` (git rm)
+- `lib/agents/continue.plugin.json` (git rm)
+- `docs/research/agent-instructions-sync.md` (git rm)
+- `docs/README-v1.3.md` (git rm)
+- `.agent/` (rm -rf, untracked)
+- `.agents/` (rm -rf, untracked)
+
+**What to implement**: Remove all files that are entirely about dropped agents. Git-tracked files use `git rm`. Untracked/gitignored directories use `rm -rf`.
+
+**TDD steps**:
+1. Write test: `test/cleanup/dropped-agent-files.test.js` — assert none of these files/dirs exist
+2. Run test: fails (files still exist)
+3. Implement: `git rm` tracked files, `rm -rf` untracked dirs
+4. Run test: passes
+5. Commit: `fix: remove dropped-agent files — aider config, continue plugin, stale research docs`
+
+**Expected output**: All 6 paths gone.
+
+---
+
+## Task 2: Remove dropped-agent code from packages/skills
+
+**File(s)**:
+- `packages/skills/src/lib/agents.js` — remove aider, antigravity, continue, windsurf entries
+- `packages/skills/src/commands/sync.js` — remove Aider updateAiderConfig(), fix help text
+- `packages/skills/test/agents.test.js` — remove Aider/Continue detection tests
+- `packages/skills/test/sync.test.js` — remove Aider sync test
+
+**What to implement**: Remove all code paths, agent entries, and tests for the 4 dropped agents from the skills package.
+
+**TDD steps**:
+1. Write test: `test/cleanup/dropped-agent-code.test.js` — grep these files for dropped agent names, assert zero matches
+2. Run test: fails (references still exist)
+3. Implement: edit all 4 files
+4. Run test: passes
+5. Run existing test suite: `bun test` — verify no regressions
+6. Commit: `fix: remove dropped-agent code from skills package — aider, antigravity, continue, windsurf`
+
+**Expected output**: No dropped-agent names in skills package source or tests.
+
+---
+
+## Task 3: Remove dropped-agent code from bin/forge.js and lib/
+
+**File(s)**:
+- `bin/forge.js` — remove Continue setup function (~40 lines), continueFormat references
+- `bin/forge-cmd.js` — remove "OpenSpec" from plan description
+- `lib/project-discovery.js` — remove Aider detection logic
+
+**What to implement**: Remove all dead code paths for dropped agents from the CLI and lib modules.
+
+**TDD steps**:
+1. Write test: `test/cleanup/dropped-agent-cli.test.js` — grep these files for dropped agent names, assert zero matches (excluding comments about removal)
+2. Run test: fails
+3. Implement: edit all 3 files
+4. Run test: passes
+5. Run `bun test` — verify no regressions
+6. Commit: `fix: remove dropped-agent code from CLI — continue setup, aider detection, openspec ref`
+
+**Expected output**: No dropped-agent function calls in CLI code.
+
+---
+
+## Task 4: Clean dropped-agent references from docs
+
+**File(s)**:
+- `docs/EXAMPLES.md` — delete Example 4 (OpenSpec-based), fix `/research` → `/plan` in examples 1,2,3,5
+- `docs/TOOLCHAIN.md` — remove Windsurf mention, Continue MCP setup
+- `docs/AGENT_INSTALL_PROMPT.md` — remove Continue detection
+- `docs/research/agent-permissions.md` — remove Antigravity/Aider rows from tables
+- `docs/research/dependency-chain.md` — fix 1 Continue reference
+- `docs/research/test-environment.md` — fix 1 Continue reference
+- `lib/agents/README.md` — remove Windsurf, Antigravity, Aider rows
+
+**What to implement**: Fix each doc per the design doc rules — delete sections that are fundamentally about dropped agents, fix minor references in docs that are otherwise valid.
+
+**TDD steps**:
+1. Write test: `test/cleanup/dropped-agent-docs.test.js` — grep all doc files for dropped agent names, assert zero matches (allow historical mentions in design docs and CHANGELOG)
+2. Run test: fails
+3. Implement: edit all files
+4. Run test: passes
+5. Commit: `docs: remove dropped-agent references from docs — antigravity, windsurf, aider, continue`
+
+**Expected output**: No misleading dropped-agent references in active docs.
+
+---
+
+## Task 5: Clean package.json, CLAUDE.md, QUICKSTART.md, .gitignore, .forge/
+
+**File(s)**:
+- `package.json` — description: "7-stage" + 8 agents only, remove dropped keywords
+- `CLAUDE.md` — remove Continue MCP reference
+- `QUICKSTART.md` — remove Windsurf from examples
+- `.gitignore` — remove `.agents/`, `.agent/`, `.aider/skills/`, `.continue/skills/`, `.windsurf/skills/`
+- `.forge/pr-body.md` — remove Aider, Antigravity references
+- `test-env/validation/agent-validator.test.js` — remove aider from list
+
+**What to implement**: Fix all remaining config and metadata files.
+
+**TDD steps**:
+1. Write test: `test/cleanup/dropped-agent-config.test.js` — assert package.json description says "7-stage", keywords don't include dropped agents, .gitignore doesn't have dropped entries
+2. Run test: fails
+3. Implement: edit all files
+4. Run test: passes
+5. Commit: `fix: clean dropped-agent refs from package.json, gitignore, quickstart, config`
+
+**Expected output**: All metadata accurate. `bun test` passes.
+
+---
+
+## Task 6: Fix plugin catalog capability flags
+
+**File(s)**:
+- `lib/agents/cursor.plugin.json` — `commands: true`, add `"commands": ".cursor/commands"` to directories
+- `lib/agents/cline.plugin.json` — `commands: true`, add `"workflows": ".clinerules/workflows"` to directories
+- `lib/agents/copilot.plugin.json` — `commands: true`
+- `lib/agents/kilocode.plugin.json` — `commands: true`
+- `lib/agents/codex.plugin.json` — `commands: true`, add `"skills": ".codex/skills"` to directories
+- `lib/agents/claude.plugin.json` — `hooks: true`
+
+**What to implement**: Update each plugin.json to reflect actual capabilities. These flags affect `forge setup` output.
+
+**TDD steps**:
+1. Write test: `test/cleanup/plugin-catalog.test.js` — assert each supported agent has `commands: true`, correct directories, no `continue` plugin exists
+2. Run test: fails
+3. Implement: edit 6 plugin.json files
+4. Run test: passes
+5. Run `bun test` — verify no regressions (existing plugin tests may need updates)
+6. Commit: `fix: update plugin catalog — correct capability flags for all 8 supported agents`
+
+**Expected output**: All 8 plugins have accurate capability flags.
+
+---
+
+## Task 7: Build `forge check-agents` CLI command
+
+**File(s)**:
+- `scripts/check-agents.js` (new) — delegates to `sync-commands.js --check` + validates plugin catalog
+- `test/scripts/check-agents.test.js` (new)
+
+**What to implement**: CLI command that:
+1. Runs `syncCommands({ check: true })` to verify all agent command files are in sync
+2. Reads `lib/agents/*.plugin.json` to verify each agent with `commands: true` has its command directory populated
+3. Reports: missing files, out-of-sync files, stale files
+4. Exits 0 if all clean, non-zero if issues found
+
+Keep it simple — reuse the existing sync infrastructure rather than reimplementing file checks.
+
+**TDD steps**:
+1. Write test: `test/scripts/check-agents.test.js` — happy path (all files present → exit 0), missing file (→ exit non-zero), stale file detection
+2. Run test: fails (file doesn't exist)
+3. Implement: `scripts/check-agents.js`
+4. Run test: passes
+5. Run manually: `node scripts/check-agents.js` → "All agent command files are in sync."
+6. Commit: `feat: add forge check-agents CLI — validates agent configs are complete and in sync`
+
+**Expected output**: `node scripts/check-agents.js` exits 0 on current repo.
+
+---
+
+## Task 8: Integration validation + design doc update
+
+**File(s)**: No new files.
+
+**What to implement**:
+1. Run full test suite: `bun test` — all pass
+2. Run `node scripts/sync-commands.js --check` — in sync
+3. Run `node scripts/check-agents.js` — all clean
+4. Grep entire codebase for dropped agent names — zero hits in active code/docs
+5. Update design doc status: "Active" → "Complete"
+6. Update original design doc (2026-03-04) status: "Draft" → "Superseded by 2026-03-15-agent-command-parity-v2-design.md"
+
+**TDD steps**:
+1. Run all checks listed above
+2. Verify each passes
+3. Commit: `docs: mark agent-command-parity design docs as complete`
+
+**Expected output**: All green. forge-2w3 ready to close after merge.
+
+---
+
+## Ordering Summary
+
+| # | Task | Blocks | Parallelizable |
+|---|------|--------|----------------|
+| 1 | Delete dropped-agent files | — | Yes (with 2-5) |
+| 2 | Remove dropped code from skills package | — | Yes (with 1,3-5) |
+| 3 | Remove dropped code from CLI/lib | — | Yes (with 1,2,4,5) |
+| 4 | Clean docs | — | Yes (with 1-3,5) |
+| 5 | Clean config/metadata | — | Yes (with 1-4) |
+| 6 | Fix plugin catalog | — | Yes (with 1-5) |
+| 7 | Build check-agents CLI | 6 (reads plugin catalog) | After task 6 |
+| 8 | Integration validation | All above | Last |
+
+Tasks 1-6 are independent and can be parallelized.
+Task 7 depends on task 6 (needs correct plugin catalog).
+Task 8 is the final integration check.