.gitignore

# Dependencies
# No trailing slash: also match worktree per-package node_modules symlinks (SMI-5124)
node_modules
.pnpm-store/

# Ruflo AgentDB runtime persistence (hive-mind state)
agentdb.rvf
agentdb.rvf.lock

# SMI-4417: doc-retrieval-mcp — local-only .rvf corpus index + metadata.
# Never committed, never built in CI. .rvf is untracked, so smudge/clean
# filters never run on it — no .git-crypt-ignore entry is needed.
.ruvector/
*.rvf

# SMI-4680: stale CycloneDX SBOM at repo root (deleted; zero consumers
# verified via grep). On-demand SBOM generation lives in
# packages/mcp-server/src/tools/compliance-tools.ts (MCP tool output).
# If a static SBOM is needed again, regenerate via:
#   docker exec skillsmith-dev-1 npx @cyclonedx/cyclonedx-npm \
#     --output-format json --output-file sbom.json
sbom.json
*.cdx.json

# Build outputs
dist/
*.tsbuildinfo
.turbo/
.astro/

# TypeScript build artifacts in source dirs
packages/**/*.js
packages/**/*.js.map
packages/**/*.d.ts
packages/**/*.d.ts.map
# Allow config files (not build artifacts)
!packages/**/vitest.config.ts
# Allow hand-written type declarations (not generated)
!packages/**/src/types/*.d.ts
# SMI-2742: Ambient module declaration for fts5-sql-bundle
!packages/core/src/db/drivers/fts5-sql-bundle.d.ts
!packages/**/eslint.config.js
!packages/**/astro.config.mjs
# SMI-4904: custom ESLint rules (hand-written, not build artifacts)
!packages/**/eslint-rules/**/*.js
vitest.config.js
vitest.config.js.map
vitest.config.d.ts
vitest.config.d.ts.map

# Coverage
coverage/

# IDE
.idea/
.vscode/
!packages/vscode-extension/.vscode/
packages/vscode-extension/.vscode/*
!packages/vscode-extension/.vscode/launch.json
*.swp
*.swo
.DS_Store

# Environment
.env
.env.local
.env.*.local
# Explicitly allow environment templates (should be committed)
!.env.example
!.env.schema

# Logs
logs/
*.log
!packages/doc-retrieval-mcp/eval/.signatures.log
npm-debug.log*

# Testing
.nyc_output/

# OS
Thumbs.db

# Misc
*.bak
*.tmp

# Coordination artifacts
.claude-flow/
.swarm/
.agentic-qe/

# Confidential business documents
docs/IP/
docs/licensing/
docs/valuation/

# Sensitive configuration documentation (contains IDs, secrets)
docs/config/
*-secrets.md
*-credentials.md
stripe-config*.md
.stripe-*

# Internal planning and strategy (not public)
docs/backlog/
docs/strategy/

# Development artifacts (anchored to root level)
# SMI-4396: /data/* instead of /data/ so the SMI-4396 allowlist negation below
# can re-include its one version-controlled file — gitignore forbids negating
# a child when the parent directory itself is excluded.
/data/*
!/data/skills-security-allowlist.json
# SMI-4408: indexer blocklist for non-skill repos (imported-skill ingest filter).
!/data/indexer-blocklist.json
# SMI-6441: vendored SecLists snapshot + hand-curated keeplist for the weak-
# password lexicon generator (scripts/gen-weak-password-lexicon.mjs). The
# directory itself must be un-ignored (not just its files) — see the SMI-4396
# comment above on why gitignore can't negate a child of an excluded parent.
!/data/wordlists/
/output/
test-results/
playwright-report/
blob-report/

# SMI-5668: NEEDLE-based Codex dispatch (maintainer-machine-only tooling).
# results/*.log is already covered by the blanket *.log rule above; these two
# are a defensive backstop against an accidental commit of NEEDLE's own
# bead-store SQLite/JSONL state or its per-bead traces from inside a worktree.
scripts/needle/results/
.beads/

# Ignore test databases and output in packages (generated during tests)
packages/*/data/
packages/*/output/

# SMI-4577: HNSW bench fixtures (regenerated via `npm run bench:hnsw:seed`)
packages/core/tests/embeddings/fixtures/

# Phase 4 working documents (not finalized)
docs/phase4/
docs/phase4-research/
docs/architecture/phase-4.5-large-scale-testing.md
docs/architecture/phase4-trigger-system.md
docs/retros/phase-4-product-strategy.md

# Development scripts (not production)
scripts/full-import.ts
scripts/security-scan.ts
scripts/security-scan.test.ts
scripts/utils/

# Untracked scaffolding/development directories
packages/web/
/src/
docs/site/
.env.registry
.migration-checkpoint.json
.batch-transform-checkpoint.json
.batch-transform-checkpoint.json.bak
.hive-mind/
.vercel

# DNS and Email Provider Exports
*.zone
*-dns-export.txt
skillsmith.app.txt
cloudflare-*.json
resend-*.json

# Cloudflare configuration
.cloudflare/
wrangler.toml.local

# Resend webhook secrets (beyond .env)
.resend-*

# Claude Code runtime files
.claude/memory.db
.claude/hive-mind/
.claude/scheduled_tasks.lock

# Ruflo (claude-flow) runtime DB created by mcp__ruflo__swarm_init + memory_store
ruvector.db
agentdb.rvf
agentdb.rvf.lock

docs/irap/
.worktrees/

# SMI-4649: scripts/create-worktree.sh auto-generates this with worktree-specific
# container_name + ports per worktree. Never to be committed — `git add -A` from
# any worktree could accidentally check it in and break sibling contributors.
docker-compose.override.yml

# MCP Registry tokens
.mcpregistry_*

# Blog image generation working directory
docs/articles/tmp/

# Understand Anything knowledge graphs (generated by /understand)
**/.understand-anything/

# SMI-5207 governance review L-3: scripts/indexer/smi5207-blast-radius-{weekly,
# transitions}.ts default their report output here instead of the repo root
# (CLAUDE.md "Never save working files, text/mds and tests to the root folder").
.smi5207-reports/
# SMI-6441 Wave 2: scripts/indexer/smi6441-blast-radius.ts follows the same
# convention for the same reason.
.smi6441-reports/
# SMI-5879 / SMI-6444 / SMI-6481 census, simulation, gate-check and acceptance
# evidence. These tools default-write to the repo root, and the files are
# campaign evidence referenced from Linear rather than committed artifacts.
# Left untracked they fail `format:check`, which scans the whole tree — that
# blocked a docs-only push on 2026-09-09 for reasons unrelated to the change.
#
# TWO patterns per prefix, not one wildcard. Each prefix needs the tool's own
# output (`.json`) plus the operator's pre-rerun backup of it (`.json.bak…`,
# e.g. `.json.bak-pre-smi6481` on a 169MB file that sat unmatched here).
#
# Scope limit, deliberate: ONLY the `.bak` backup convention is covered. A
# backup saved under any other suffix — `.json.gz`, `.json.old`, `.json.copy`
# — stays VISIBLE, where the old `.json*` had swallowed it. That narrowing is
# intended, not an oversight: #2795's cross-family pre-merge review raised the
# compressed case, and a grep for gzip/zlib/zstd/zip/archiver across the three
# tools found no compression path in any of them, so re-widening for a shape
# nothing produces would repeat exactly the guess this block was rewritten to
# undo. If a tool ever starts compressing, add that explicit suffix per prefix
# — never restore `.json*`.
#
# An earlier revision collapsed both into a single `.json*` and claimed
# "`.json` still anchors the match". That was WRONG, and #2789's cross-family
# pre-merge review caught it: `.json*` means "STARTS WITH `.json`", so it also
# swallowed root-level `.jsonl`, `.json5`, `.jsonnet`, `.json.ts` and
# `.json.example` — real formats someone could legitimately want to commit.
# Silently hiding a source or data file is exactly the risk this block must
# not create. Confirmed with `git check-ignore -v --no-index` in both
# directions before and after.
#
# Every pattern is root-anchored (`/`) deliberately: without the leading slash
# git matches the basename at ANY depth, which would hide a same-named fixture
# or committed evidence file under a subdirectory.
/smi5879-*.json
/smi5879-*.json.bak*
/smi6444-*.json
/smi6444-*.json.bak*
/smi6481-*.json
/smi6481-*.json.bak*