Skip to content

What's included

The big picture: Blueprint is about 140 files of process — CI, Git hooks, scripts, Claude Code agents and skills — and no application code. This page lists all of it.

Why it matters: You own everything you clone. Knowing what each part does tells you what’s safe to delete and what’s quietly holding something up.

The bottom line: Nothing here is mandatory. Delete what you don’t need — but delete it on purpose (see Removing what you don’t need).

Category Files What it gives you
Project contract AGENTS.md The agent-agnostic rules any coding agent reads: stack, commands, invariants and forbidden operations, the branch → MR → green-pipeline flow, where the gates live — see Agent portability
Claude agents .claude/agents/ Specialist reviewers and helpers — see Agents
Claude skills .claude/skills/ Slash-command workflows — see Skills
Claude hooks .claude/hooks/, .claude/settings.json Guards that block risky edits, reminders after edits, and a security gate before an MR opens
Claude rules CLAUDE.md (imports AGENTS.md on line 1), .claude/operating-rules.md, nested scripts/CLAUDE.md and ci/CLAUDE.md, .claude/rules/ (backend.md, frontend.md, tests.md, api-compat.md, docs.md, claude-config.md, ux-copy.md, specs.md), reference material in docs/harness/, global-claude-md.example (optional) The Claude Code workflow on top of the contract: skills, agents, hooks, the fast-path gate chain. Claude reads these automatically; nested files load only for their folder, and a rule file loads only when Claude works with a file matching its paths: globs
Personas .claude/personas.md The users /voc, /voc-audit, and /sunset-check role-play
Security invariants .claude/security-invariants.md Cross-surface security rules architect, rbac-check, security-review, and threat-model check against — see Security invariants registry
CI pipeline .gitlab-ci.yml, ci/*.yml Governance jobs out of the box, plus stack templates — see CI pipeline
Git hooks hooks/, scripts/setup-hooks.sh Pre-commit: secret scan, format-check, lint, typecheck. Pre-push: make pre-push-checks
Behavioral evals evals/, ci/evals.yml, scripts/check-evals.py, scripts/run-evals.sh, scripts/eval-plugin-dir.sh, scripts/eval-trend.py Planted-defect cases that score the agents and skills with claude plugin eval, with and without them loaded; a free lint on every MR and a paid weekly run. Ships in 0.8 — see Evaluating the harness
Check scripts (gates) scripts/check-*, scripts/osv-severity-gate.sh, scripts/gitleaks-precommit.sh, load-bearing.declarations, .gitleaks.toml Checks that each prove they can still fail — see Harness gates
Optional templates scripts/templates/check-served-assets.sh Copy-in smoke check that a built single-page-app image serves its own assets; a template, not a gate Blueprint runs on your code. Only its --self-test runs in CI — see Served-assets smoke test
Parallel work scripts/wt One worktree per issue, safe with several sessions at once — see Parallel work
Changelog changelog.d/, scripts/assemble-changelog.sh One-line fragments per MR, assembled at release
Release scripts/release.sh, scripts/release-notes.sh, docs/releases/ Version bump, changelog assembly, per-release notes
Templates .gitlab/issue_templates/, .gitlab/merge_request_templates/ Bug, Feature, Feedback, and Task issues; a standard MR description
Specs docs/specs/, scripts/check-spec-completeness.py, scripts/spec-rules/ Product specs, their template, and the completeness gate that decides whether a spec can become issues — see Spec-driven development
ADRs docs/adr/ Architecture Decision Records, created with /adr
Docs site website/ This site (Astro Starlight → GitLab Pages). The reset deletes it by default; --keep-website keeps a stub as a starting point for yours
Dependency updates renovate.json, .github/dependabot.yml Automated update MRs — keep one, delete the other
Setup Makefile, scripts/customize.sh, scripts/doctor.sh, scripts/gitlab-bootstrap.sh, scripts/blueprint-status make setup, make customize (setup checklist), make status (lifecycle checklist: Configure → Specify → Plan → Build → Release, ships in 0.8), make doctor (prerequisites), and /kickoff’s GitLab bootstrap (project, settings, protections, labels, seed issues)
Conventions CONTRIBUTING.md, .gitattributes, .env.example Branch and commit rules, line endings, example environment
  • No frontend? Delete ci/node.yml, the node blocks in the Makefile, frontend/, and the ux-design, ux-review, and accessibility agents. If you don’t ship a single-page-app image, also delete scripts/templates/check-served-assets.sh together with its --self-test lines in the tooling-self-tests job of .gitlab-ci.yml and in the gate-self-tests target of the Makefile; deleting only the script fails CI.
  • No HTTP endpoints? Delete the rbac-check agent and the pre-mr-security-gate.sh hook (and its PreToolUse entry in .claude/settings.json). Delete its eval cases too, evals/review/authz-detail-idor/, evals/review/grant-path-parity/, evals/review/grant-path-shared-helper/, evals/review/grant-path-clean-control/ and evals/review/rbac-clean-control/: with the agent gone they can only fail. The same holds for any agent or skill you delete. grep -rl '<name>' evals/ finds its cases.
  • Not running the behavioral evals? Delete evals/ (both of ci/evals.yml’s jobs then skip themselves), or delete ci/evals.yml, its include line in .gitlab-ci.yml, the four scripts/*eval* scripts, scripts/tests/eval-trend/, and their lines in the Makefile, tooling-self-tests and load-bearing.declarations.
  • No public API contract yet? Delete .claude/rules/api-compat.md and the Backward compatibility section of AGENTS.md.
  • No docs site? Nothing to do after the reset: it deletes website/ by default, and the website:build and pages jobs skip themselves without it. (ci/docs.yml can stay — it does nothing without a docs site.) Without the reset, delete website/ by hand.
  • No Helm chart? Leave ci/helm.yml commented out, or delete it.
  • No CI? Delete .gitlab-ci.yml and ci/. The changelog and hooks still work locally.
  • No Claude Code? Delete .claude/, evals/ and global-claude-md.example, and remove the @.claude/operating-rules.md import line from CLAUDE.md. Keep AGENTS.md: it is the whole project contract, and tools that follow the AGENTS.md convention read it directly. Keep CLAUDE.md too: scripts/gate-yield.py --check-fastpath, which CI runs, reads its fast-paths table. Everything else works on its own. What another agent does and does not get: Agent portability.
  • Different license? Replace LICENSE.
  • GitHub instead of GitLab? Move .gitlab-ci.yml to .github/workflows/ and .gitlab/ to .github/; the scripts already use gh when origin is on GitHub. Delete renovate.json and keep .github/dependabot.yml (scripts/customize.sh --reset-template --github does this).
  • GitLab (the default)? Delete .github/dependabot.yml; keep renovate.json (scripts/customize.sh --reset-template --gitlab does this).

Yes, but: before deleting a check script, look for it in load-bearing.declarations and the Makefile’s pre-push-checks list. Deleting a gate cleanly means removing its callers too.

Add What it gives you When
Per-path CODEOWNERS rules Automatic reviewer assignment by path (/kickoff already writes a root CODEOWNERS with a default owner) More than two people with distinct areas
Container image scanning CVE scanning of built images (Trivy, GitLab container scanning) Shipping containers to production
Language vulnerability jobs pip-audit / npm audit alongside osv-scan You want the ecosystem’s native advisories too
A richer MR template Migration notes, deploy impact, rollback plan Deploys with migrations or several services
Approval rules N approvals before merge, per-path overrides A formal review process
Merge trains GitLab’s server-side merge queue main breaks often from concurrent merges (/mass-merge is the client-side version)

Choosing where to deploy is a separate question — see Deployment options.

⛔ marks a skill only you can start (disable-model-invocation: true).

.
├── .claude/
│ ├── agents/ # the agents — see the Agents page
│ │ ├── accessibility.md # WCAG 2.1 AA (sonnet, read-only)
│ │ ├── architect.md # technical approach before code (opus)
│ │ ├── changelog.md # writes the changelog fragment (sonnet)
│ │ ├── completeness-check.md # fresh-agent audit of branch vs. issue, before push (opus by default; opus or sonnet by risk tier under /wave)
│ │ ├── dependency.md # license/CVE/justification before adding a package (sonnet)
│ │ ├── docs.md # writes and updates docs (sonnet)
│ │ ├── generated-artifact-check.md # stale schemas, types, SDKs (sonnet)
│ │ ├── perf-check.md # N+1, unbounded queries (sonnet, read-only)
│ │ ├── rbac-check.md # access control on endpoints (opus, read-only)
│ │ ├── regression-check.md # what this breaks that used to work (sonnet)
│ │ ├── schema-check.md # migration safety, before push (sonnet)
│ │ ├── security-review.md # OWASP Top 10 + project risks (opus; /wave runs it on sonnet)
│ │ ├── spec-review.md # semantic review of a draft spec (opus, read-only, report-only)
│ │ ├── test-scaffold.md # tests for uncovered code (sonnet)
│ │ ├── threat-model.md # trust boundaries, before code (opus)
│ │ ├── ux-design.md # UI proposal, before code (opus)
│ │ └── ux-review.md # design-system compliance (sonnet, read-only)
│ ├── skills/ # the skills — see the Skills page
│ │ ├── adr/SKILL.md # /adr
│ │ ├── ci-debug/SKILL.md # /ci-debug
│ │ ├── conventions-review/SKILL.md # /conventions-review (was /review)
│ │ ├── dotplanning/ # /dotplanning ⛔
│ │ │ ├── SKILL.md
│ │ │ ├── kickoff-questions.md # the twelve kickoff questions
│ │ │ └── html-output-spec.md # report layout
│ │ ├── fix-mr/SKILL.md # /fix-mr ⛔
│ │ ├── import-design/SKILL.md # /import-design
│ │ ├── import-spec/SKILL.md # /import-spec ⛔ (deprecated alias of /spec import)
│ │ ├── incident-postmortem/SKILL.md # /incident-postmortem
│ │ ├── kaizen/SKILL.md # /kaizen
│ │ ├── kickoff/ # /kickoff ⛔
│ │ │ ├── SKILL.md
│ │ │ └── stack-map.md # stack and answer → file edits
│ │ ├── mass-merge/ # /mass-merge ⛔
│ │ │ ├── SKILL.md
│ │ │ └── failure-modes.md # failure modes and recovery
│ │ ├── memory-audit/SKILL.md # /memory-audit ⛔
│ │ ├── mr/SKILL.md # /mr ⛔
│ │ ├── next/SKILL.md # /next
│ │ ├── pre-release/SKILL.md # /pre-release
│ │ ├── release/SKILL.md # /release ⛔
│ │ ├── spec/ # /spec ⛔
│ │ │ ├── SKILL.md # commands, bare dispatch by status, aliases
│ │ │ └── reference/
│ │ │ ├── interview.md # new, resume, audit, revise
│ │ │ └── import.md # import, --dry-run, re-import diff
│ │ ├── spec-intake/SKILL.md # /spec-intake (deprecated alias of /spec)
│ │ ├── sunset-check/SKILL.md # /sunset-check
│ │ ├── tracker-hygiene/SKILL.md # /tracker-hygiene
│ │ ├── voc/SKILL.md # /voc
│ │ ├── voc-audit/SKILL.md # /voc-audit ⛔
│ │ └── wave/SKILL.md # /wave ⛔ (was /batch)
│ ├── hooks/
│ │ ├── context-nudge.sh # Stop hook: tells you once when context passes a threshold (ships in 0.8)
│ │ ├── on-stop.sh # Stop hook: blocks until `make test-scoped` ran (silent on the stub)
│ │ ├── post-edit-checks.sh # reminders after an edit
│ │ ├── post-mr-cost.sh # PostToolUse: posts the MR's cost record after `glab mr create`
│ │ ├── pre-mr-security-gate.sh # stops `glab mr create` until security reviews ran
│ │ ├── pre-tool-agent-budget.sh # PreToolUse: stops a /wave implementer past 120 tool calls (ships in 0.8)
│ │ ├── pre-tool-safety.sh # blocks lock-file and migration edits
│ │ └── session-start-status.sh # SessionStart: one progress line (ships in 0.8)
│ ├── cost-ledger.json # cost ledger setting: on or off (write it with scripts/cost-ledger)
│ ├── model-prices.json # dated list-price table scripts/est-cost.py reads for est-cost-usd:
│ ├── kaizen-declined.json # /kaizen proposals you've turned down
│ ├── operating-rules.md # gate reference + cross-cutting rules, imported by CLAUDE.md
│ ├── personas.md # your users (replace the examples)
│ ├── rigor.json # the rigor floor: paths whose MRs run at strict or strict-light (add your migrations and auth code)
│ ├── rules/ # path-scoped rules: backend, frontend, tests, api-compat, docs, claude-config, ux-copy, specs
│ ├── security-invariants.md # cross-surface security rules (replace the examples)
│ └── settings.json # hooks, allowed and denied commands
├── .github/dependabot.yml # GitHub dependency updates (delete on GitLab)
├── .gitlab/
│ ├── issue_templates/ # Bug.md, Feature.md, Feedback.md, Task.md
│ └── merge_request_templates/ # Default.md
├── changelog.d/README.md # fragment naming guide
├── ci/
│ ├── CLAUDE.md # rules for editing CI (imports scripts/CLAUDE.md)
│ ├── cloudflare-deploy.yml # marketing-site deploy template (opt-in, fully commented out)
│ ├── docker.yml # kaniko build, compose project names
│ ├── docs.yml # internal docs link check (included by default)
│ ├── go.yml # golangci-lint, go test, govulncheck
│ ├── helm.yml # chart registry drift, install/upgrade drills
│ ├── mutation.yml # optional nightly mutation gate (mutmut example, off by default)
│ ├── node.yml # eslint, vitest, license check, added-file coverage
│ ├── python.yml # ruff, pytest, license check, app-aware gates, fuzz, load test
│ ├── sonar-complexity.yml # cognitive-complexity tier (off by default)
│ ├── sonar-exclusions.yml # Sonar ignore-criteria rule-key check (off by default)
│ └── trivyignore-expiry.yml # .trivyignore expiry-date check (off by default)
├── docs/
│ ├── adr/ # ADR format guide, seed ADR, and Blueprint's own ADRs
│ ├── harness/ # reference moved out of CLAUDE.md and operating-rules.md: Claude Code config, branching, agent workflow, milestone labels, the rigor floor, VoC, memory and mass-merge rules
│ ├── specs/ # product specs, TEMPLATE.md and a sample, checked by check-spec-completeness.py
│ └── releases/ # Blueprint's own release notes (replace with yours)
├── hooks/
│ ├── pre-commit # secret scan, format-check, lint, typecheck (each if defined)
│ └── pre-push # make pre-push-checks
├── scripts/
│ ├── CLAUDE.md # rules for writing check scripts
│ ├── lib/
│ │ ├── codeowners-locations.sh # where a CODEOWNERS file counts (customize.sh and the /kickoff writer share it)
│ │ ├── git-hooks-dir.sh # finds the real hooks folder (works in worktrees)
│ │ └── git-ignored.sh # lets checks skip ignored files, like CI's clean clone
│ ├── spec-rules/ # data for the spec gate: section aliases, vague terms, verdict schema
│ ├── tests/ # tests for wt, setup, onboarding, the security hook, app-aware gates
│ ├── adr-accepted-issue-sweep.py # open issues written before the ADR they cite (or the spec D-nn)
│ ├── assemble-changelog.sh # fragments → CHANGELOG.md
│ ├── check-added-files-covered.mjs # a new source file missing from the coverage report
│ ├── check-artifact-assertions.sh # tag jobs assert their artifacts aren't empty
│ ├── check-chart-registry.sh # the published Helm chart installs as documented
│ ├── check-cognitive-complexity.py # complexity tier for script entry points
│ ├── check_mutation_score.py # mutation gate: merge shards, score floor, fails closed
│ ├── check-compose-project-names.sh # each compose file pins its own project name
│ ├── check-docs-internal-links.py # docs links, anchors, and assets resolve
│ ├── check-edition-boundary.sh # the open-source tree never names the enterprise package (open core)
│ ├── _paths.py # shared path-containment helper (resolve_within)
│ ├── mutmut_export.py # mutmut 2.x cache to shard JSON (mutation gate adapter)
│ ├── mutmut_config.py.example # shard hook for the mutation gate
│ ├── check-gate-selftest-parity.sh # every gate self-tests in its own CI job
│ ├── check-issue-collision.sh # blocks a duplicate MR for an issue
│ ├── wave-preflight.py # /wave: is this issue already done on main?
│ ├── wave-resume.py # /wave: classify leftover claims and worktrees
│ ├── cost-ledger # reads and writes the cost ledger setting (on, off, status)
│ ├── orchestrator-tokens.py # /wave: sums the orchestrating session's token usage for a branch
│ ├── subagent-tokens.py # /wave: sums the branch's subagents' token usage, descendants included
│ ├── est-cost.py # /wave: prices the branch's token usage at list prices (est-cost-usd:)
│ ├── cost-table.py # renders and posts the numbered ## Cost of this MR comments
│ ├── post-mr-cost.py # the only code that measures and posts them (hook, --round, --measure)
│ ├── check-load-bearing.sh # declared one-line guards still exist
│ ├── check-memory-index.sh # Claude memory index size and dead links
│ ├── check-migration-numbering.sh # migration numbers that collide on the merged tree
│ ├── check-mr-followups.sh # a declared follow-up must cite an open issue
│ ├── check-spec-trace.sh # imported specs and their issues must trace both ways
│ ├── check-gate-ledger.py # MR `## Gates` outcomes must be parseable
│ ├── check-rigor-floor.py # an MR touching the rigor floor must record strict or strict-light and its gate lines
│ ├── check-ci-include-all.py # CI files parse with every include enabled
│ ├── check-nul-bytes.sh # no NUL bytes in text files
│ ├── check-prepush-parity.sh # every CI gate can also run before push
│ ├── check-release-pipeline.py # a tag can't publish unless every job at its commit passed
│ ├── check-security-deferred.sh # every issue that ever carried security::deferred is resolved
│ ├── check-spec-completeness.py # a spec under docs/specs/ is complete enough to become work
│ ├── check-single-fastpath-table.sh # one fast-paths table; no rule heading in both AGENTS.md and CLAUDE.md
│ ├── check-serializer-ts-parity.py # backend schema and TypeScript types agree
│ ├── check-shellcheck.sh # every shell script passes shellcheck
│ ├── check-sigpipe-readers.sh # no early-exit reader behind a pipe
│ ├── check-shell-portability.sh # no BSD-only or GNU-only shell idiom (stat/sed -i/date/grep -P ...)
│ ├── check-cc-claims.sh # Claude Code claims need a doc link + checked date
│ ├── check-skill-names.sh # no skill shadows a Claude Code built-in name or alias
│ ├── check-commands-doc.py # every skill has a commands.md entry with Before this:/Next:
│ ├── check-agent-contract.sh # agent "Do not use" clauses, one severity scale
│ ├── check-claudemd-size.sh # no CLAUDE.md over 200 lines (allowlist: claudemd-size-allowlist.txt), byte budget: claudemd-byte-budget.txt
│ ├── check-stale-references.sh # stale-reference rules (see harness-gates)
│ ├── check-tag-only-jobs.sh # tag-only CI jobs changed since the last tag
│ ├── check-version-lockstep.py # every version-bearing file agrees
│ ├── check-yaml-duplicate-keys.py # no repeated key in .gitlab-ci.yml or ci/*.yml
│ ├── check-bare-issue-refs.py # no bare #NNN Blueprint issue reference in shipped files
│ ├── check-harness-reference.py # every hook, scripts/wt subcommand and make target has a reference-page entry
│ ├── check-docs-stale.py # no doc quotes a make target, script or flag that no longer exists
│ ├── check-ws-event-reachability.py # published real-time events are actually reachable
│ ├── blueprint-status # make status: the lifecycle checklist (Configure → Specify → Plan → Build → Release) and --json / --require for skills
│ ├── ci-assert-artifacts.sh # helper: fail a publish job with an empty artifact
│ ├── customize.sh # setup checklist; --init starts the repo
│ ├── detect-stack.py # /kickoff's stack detection: proposes Makefile recipes, the CI include and rules globs as a diff; writes only on a confirmed plan id
│ ├── doctor.sh # prerequisite check
│ ├── gate-yield.py # per-gate yield, escapes and cost from the MR ledger; --by-model: yield per gate and model; --spec-trace: spec-intake yield
│ ├── gitleaks-precommit.sh # secret scan for the pre-commit hook
│ ├── gitlab-bootstrap.sh # /kickoff's GitLab project, settings, protections, labels and seed issues
│ ├── helm-install-drill.sh # install + upgrade the chart on a real kind cluster
│ ├── kickoff-state.py # /kickoff's saved answers per round (.claude/state/kickoff.json), so a re-run resumes
│ ├── mr-verdict.py # one MR's verification verdict from its Gates/Requirements
│ ├── nightly_load_test.py # p95 latency budgets
│ ├── osv-severity-gate.sh # fail on dependency CVEs above a threshold
│ ├── release-notes.sh # docs/releases/<version>.md
│ ├── release.sh # version bump + changelog assembly
│ ├── schemathesis_hooks.py.example # starter hooks for the schema fuzz job
│ ├── security-deferred-verdict.sh # accepted-risk note verdict: regex, calendar, expiry
│ ├── setup-hooks.sh # installs hooks/ into Git
│ ├── templates/
│ │ └── check-served-assets.sh # optional template, not a gate: an SPA image serves its own assets (self-test runs in CI)
│ ├── write-ownership-files.sh # /kickoff's CODEOWNERS, SECURITY.md, support line and LICENSE copyright line
│ ├── write-edition-boundary.sh # /kickoff's Open core follow-up: writes edition-boundary.conf
│ └── wt # worktree helper: issue check-out lock, private `wt stash`
├── website/ # this docs site
├── .env.example
├── .gitattributes # line endings, merge strategies, binary detection
├── .gitignore
├── .gitlab-ci.yml # governance CI + stack includes
├── .gitleaks.toml # secret-scan config and allowlist
├── CHANGELOG.md
├── AGENTS.md # project contract for any coding agent
├── CLAUDE.md # Claude Code workflow; imports AGENTS.md on line 1
├── CONTRIBUTING.md # contributor workflow
├── LICENSE # Apache 2.0 (change if you like)
├── Makefile # setup, customize, doctor, lint, test, build, checks
├── README.md
├── global-claude-md.example # optional personal-preferences template
├── load-bearing.declarations # one-line guards that must not disappear
├── nightly-load-test-budgets.json # latency budgets for the load test
└── renovate.json # GitLab dependency updates (delete on GitHub)