cqs-bootstrap
Skill18 starsChanged 5 months agoMITView on GitHub
One-command setup for cqs in a new project — skills, tears infrastructure, CLAUDE.md, init, index.
In this repository22 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- archeoSkill
- auditSkill
- before-editSkill
- boundary-probeSkill
- check-my-workSkill
- cqs-batchSkill
- cqs-planSkill
- cqsSkill
- cqs-verifySkill
- docs-reviewSkill
- groom-notesSkill
- idleSkill
- investigateSkill
- landSkill
- migrateSkill
- prSkill
- recall-gateSkill
- reindexSkill
- releaseSkill
- troubleshootSkill
- update-tearsSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 2 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
Check if files already exist — don't overwrite
- Very specific: line 23
Create docs/ directory if needed
- Very specific: line 124
Run cqs init (creates .cqs/ directory with database)
- Very specific: line 125
Run cqs index (indexes all source files + notes)
- Very specific: line 126
Verify with cqs stats — should show chunk count > 0
- Very specific: line 130
Add .cqs/ to .gitignore if not already present (the index database is local, not shared)
- Distinctive: line 134
If CLAUDE.md exists, append the cqs sections below.
- Very specific: line 257
Run cqs stats to confirm indexing worked
- Very specific: line 258
Test a search: cqs "main entry point" --json (should return results)
- Very specific: line 266
If cqs binary isn't found, stop and tell the user to install it first
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Bootstrap Project
Fully self-contained setup for cqs in a new project. After running this, the project has semantic search, notes, continuity tracking, and all skills working.
Prerequisites
cqsbinary installed (check withwhich cqs)- Project has a git repo initialized
- Running inside Claude Code in the target project directory (or pass project path as argument)
Process
Phase 1: Tears Infrastructure
- Check if files already exist — don't overwrite
- Create
docs/directory if needed - Write each file:
docs/notes.toml
# Notes - unified memory for AI collaborators
# Surprises (prediction errors) worth remembering
# sentiment: DISCRETE values only: -1, -0.5, 0, 0.5, 1
# -1 = serious pain, -0.5 = notable pain, 0 = neutral, 0.5 = notable gain, 1 = major win
[[note]]
sentiment = 0
text = "Project bootstrapped with cqs tears infrastructure."
mentions = ["docs/notes.toml", "PROJECT_CONTINUITY.md"]PROJECT_CONTINUITY.md
# Project Continuity
## Right Now
(active task - update when starting something)
### What happened this session
### Tracked issues
### What's next
## Parked
(threads to revisit later)
## Open Issues
## Architecture
(version, languages, tests, key tech decisions)ROADMAP.md
# Roadmap
<!-- When this grows past ~100 lines, archive completed phases to docs/roadmap-archive.md -->
## Current Phase
### Done
- [ ] ...
### Next
- [ ] ...Phase 2: Skills
- Create
.claude/skills/directory - Copy all portable skills from
/mnt/c/Projects/cqs/.claude/skills/:cqs— unified CLI dispatcher (search, graph, quality, notes, infrastructure — all subcommands)cqs-bootstrap— this skill (for nested projects)cqs-batch— batch mode: persistent Store + Embedder, stdin commands, JSONL output, pipeline syntaxcqs-plan— task planning with scout data + task-type templatescqs-verify— exercise all command categories, catch regressionsbefore-edit— run before modifying a function (impact + test-map + explain → checklist)investigate— run before starting a task (scout + gather → implementation brief)check-my-work— run after changes (review diff → risk checklist)update-tears— session state capturegroom-notes— note cleanupdocs-review— check docs for stalenessreindex— rebuild index with statstroubleshoot— diagnose common cqs issuesmigrate— handle schema version upgradesland— land a ready branch: push, PR, pinned CI watch, squash-merge, issue verification, cleanupidle— the idle work loop: enumerate issues + triage rows, classify, pick, dispatcharcheo— comment-archaeology sweep: TODO/FIXME/deferral language → issues or trivial fixesrecall-gate— retrieval recall gate with dead-gold triage + binary A/B regression testred-team— adversarial testing against cqsaudit— code audit with parallel category agentspr— WSL-safe PR creation (always --body-file)release— version bump, changelog, publish, GitHub release
- Create
.claude/agents/directory - Copy all agent definitions from
/mnt/c/Projects/cqs/.claude/agents/:investigator— pre-implementation investigation (scout + gather → brief)code-reviewer— diff review (review + impact → risk report)test-finder— test coverage lookup (test-map + impact → coverage report)implementer— implementation with cqs checkpoints (scout before, review after)lane-implementer— implementation lane with the full gate battery baked in (private target dir, all-targets clippy, targeted tests, provenance lint, commit-don't-push)explorer— codebase exploration via cqs semantic searchauditor— code audit agent with cqs tools built inseam-auditor— composition adversary (finds two correct units whose join lies)property-auditor— property-based testing lane (proptest: invariant + generator finds the input examples miss)interleaving-auditor— concurrency adversary (loom/stress: finds the schedule that breaks a shared invariant)
Phase 3: cqs Init & Index
- Run
cqs init(creates.cqs/directory with database) - Run
cqs index(indexes all source files + notes) - Verify with
cqs stats— should show chunk count > 0
Phase 4: .gitignore
- Add
.cqs/to.gitignoreif not already present (the index database is local, not shared)
Phase 5: CLAUDE.md Integration
- If CLAUDE.md exists, append the cqs sections below. If it doesn't exist, create it with these sections plus a basic header.
Check for existing sections first — don't duplicate if the user already has cqs config in their CLAUDE.md.
Sections to add to CLAUDE.md:
## Read First
* `PROJECT_CONTINUITY.md` -- what's happening right now
* `docs/notes.toml` -- observations indexed by cqs (warnings, patterns)
* `ROADMAP.md` -- what's done, what's next
## Skills
Project skills in `.claude/skills/`. Use `/skill-name` to invoke.
Skills are auto-discovered — they appear in `/` autocomplete automatically.
## Code Intelligence — When to Use What
**Use these cqs commands at the right moments.** They replace multiple manual searches with a single call. Workflow skills are even easier:
- `/before-edit <function>` — impact + tests + callers → modification checklist
- `/investigate <task>` — scout + gather → implementation brief
- `/check-my-work` — review current diff → risk assessment
### Before modifying a function:
```bash
cqs impact <function_name> --jsonBefore writing tests:
cqs test-map <function_name> --jsonBefore starting any implementation task:
cqs scout "task description" --jsonExploring unfamiliar code:
cqs onboard "concept" --json
cqs gather "query" --jsonSearching (use instead of grep/glob):
cqs "search query" --json
cqs "function_name" --name-only --json
cqs "query" --include-type function --json # filter by chunk type
cqs "query" --exclude-type test --json # exclude chunk types
cqs read <path>
cqs read --focus <function>Full command reference
cqs scout "description"— pre-investigation dashboard: search + callers/tests counts + staleness + notescqs task "description"— one-shot implementation context: scout + code + impact + placement + notescqs explain <fn>— function card: signature, callers, callees, similarcqs callers <fn>/cqs callees <fn>— call graph navigation (add--cross-projectfor multi-repo)cqs deps <type>— type dependenciescqs impact <fn> --type-impact— impact analysis with type dependenciescqs similar <fn>— find similar codecqs related <fn>— co-occurrence: shared callers, callees, typescqs where "description"— placement suggestioncqs trace <source> <target>— shortest call pathcqs context <file>— module overviewcqs impact-diff [--base REF]— diff-aware impactcqs review— diff review with risk scoringcqs ci [--base REF] [--gate high|medium|off]— CI gatecqs blame <fn>— semantic git blamecqs dead— find dead codecqs health— codebase quality snapshotcqs stale— check index freshnesscqs notes add/update/remove/list— manage project notescqs stats— index statisticscqs batch— batch mode with pipeline syntaxcqs brief <file>— one-line-per-function summarycqs affected [--base REF]— diff -> changed functions -> callers -> testscqs neighbors <fn>— embedding-space nearest neighborscqs plan "description"— task planning with templatescqs test-map <fn>— find tests that exercise a functioncqs doctor [--fix]— check model, index, hardwarecqs diff <ref>/cqs drift <ref>— semantic diff/driftcqs suggest— auto-suggest notes from patternscqs chat— interactive REPLcqs audit-mode on/off— toggle audit modecqs gc— clean stale index entries
Run cqs watch --serve in a separate terminal (or a systemd user service) to keep the index fresh AND serve fast daemon queries; cqs index for one-time refresh. CQS_NO_DAEMON=1 forces CLI mode.
Result trust — what you can act on without re-reading
cqs results carry calibration metadata. Act on it directly instead of paying a defensive re-read tax for what's already answered:
- Edge provenance (
edge_kindoncallers/callees/impactentries):callis syntactic ground truth;macro_heuristic/doc_referenceare guesses (a token-tree match or a mention in prose, not a real call). Weight a heuristic edge lower than acalledge without opening the file. Absent ⇒call. - Dead verdicts (
cqs dead --verdict): the tool classifies its own output —test-only/low-confidence-live/known-gap/dead. Trust the verdict; onlydeadis a confident absence claim. UseType::methodqualified queries to disambiguate same-named functions. - Ranking provenance (
rank_signalsper JSON result): why a hit ranked —dense(concept match) /fts(literal-string match) /name_match/note_boost. A concept match justifies reading the chunk; a string match on a conceptual query is a known false friend; anote_boostis a prior opinion, not evidence. Suppress with--no-rank-signalson tight-budget calls.
When working inside a git worktree, cqs search (and a bare query) is overlay-aware by default — it reflects your uncommitted/committed edits, not the parent branch's index (opt out with --no-overlay or CQS_WORKTREE_OVERLAY=0). The overlay is search-only: graph and scout commands (callers, impact, scout, gather, dead, …) still serve the parent index in a worktree, so re-read the actual files before acting on their output there.
Audit Mode
Before audits or fresh-eyes reviews:
cqs audit-mode on to exclude notes and force direct code examination.
After: cqs audit-mode off or let it auto-expire (30 min default).
Continuity (Tears)
"Update tears" = capture state before context compacts.
PROJECT_CONTINUITY.md-- right now, parked, blockers, open questions, pendingdocs/notes.toml-- observations with sentiment (indexed by cqs)
Use cqs notes add to add notes — it is available immediately. Direct file edits require cqs index to sync to SQLite.
Sentiment is DISCRETE — only 5 valid values: -1, -0.5, 0, 0.5, 1
### Phase 6: Verify
13. Run `cqs stats` to confirm indexing worked
14. Test a search: `cqs "main entry point" --json` (should return results)
15. Report summary: files created, chunks indexed, skills installed
## Rules
- **Never overwrite** existing files — skip with a message
- **Append, don't replace** CLAUDE.md content
- **Ask before** modifying `.gitignore` if it has complex rules
- If `cqs` binary isn't found, stop and tell the user to install it firstDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

