Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/w00fx/spec-anchored-agentic-development/prepgit clone --depth 1 https://github.com/w00fx/spec-anchored-agentic-developmentWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00079 | $0.02187 |
| Opus 5 | $0.00039 | $0.01094 |
| Sonnet 5 | $0.00016 | $0.00437 |
| Haiku 4.5 | $0.00008 | $0.00219 |
Grade A, and why
prep scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 171 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Prepare this repository so the implement skills, the reviewer, and CI have the verification infrastructure they expect. Additive and brownfield-safe: read before writing, extend instead of replacing, flag anything you had to work around. No application code changes.
Phase 0 — Detect
Identify: stacks present (TypeScript? Go? both?), layout (monorepo? capability folders?), existing test runners and configs, existing Makefile / package scripts / justfile, existing CI workflows, current coverage if measurable. Report the findings in one short block before changing anything.
Phase 1 — The commands interface (the contract)
The harness learns how to verify from three declared commands. The tool is free (Make, just, npm scripts — follow repo convention); the interface is fixed:
check— lint + typecheck + tests, whole repo.check-<capability>— the same, scoped to one capability folder (the chunk loop's command; a pattern — one target per capability).golden— the reference-value tests only, readingspecs/<capability>/tables/.
Create or extend them. Then declare them in the root AGENTS.md
under ## Commands (create it if absent, with a one-line CLAUDE.md
beside it containing @AGENTS.md — the dual-harness convention).
Phase 2 — The metric classes (stack-agnostic)
The gauntlet is defined by metric classes, not tools. Every class below must have a living instance when prep finishes, whatever the stack: the named tools are the instances for TypeScript / Go; on any other stack, find the class equivalent. Respect existing instances. A class with no instance at the end is a named blocker — never a silent absence; a class the stack genuinely lacks (e.g. race detection) is declared N/A in the report — declared, not skipped:
- Lint + typecheck: ESLint +
tsc --noEmit(TS);golangci-lintorgo vetas the floor (Go); the stack's equivalents elsewhere. - Tests + coverage: the repo's runner with a coverage reporter
(Vitest / Jest;
go test ./...with-coverprofile). - Concurrency / race detection where the stack offers it
(
go test -race); N/A-declared where it doesn't. - Secrets scan:
gitleaks— stack-independent, always present, blocking from the first run: secrets are incident-class, not ratchet-class. A true finding means rotate now; triage immediately; never grandfather a secret. (The ratchet posture applies to lint and security-analysis counts — measure first, then block-on-new — not here.) - Security static analysis:
gosec(Go); the stack's equivalent elsewhere (semgrep,npm audit,bandit, …). - Build.
- E2E / browser (web stacks): a deterministic browser suite as
check-e2e— Playwright as the instance: one scenario per UI-facing acceptance criterion, blocking in CI; N/A-declared for pure backend. The agent-facing access surface (agent-browser, guide loaded via its ownskills get corestub) is the loop's instrument, never the gate.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 171 lines · 79 tokens per session scan A 19d443791c91
prep is a command published in the GitHub repository w00fx/spec-anchored-agentic-development (5 stars, last pushed 8d ago), licensed MIT. It adds 79 tokens to every session and 2,187 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
test-pr-docker
Test PR with Docker image build workflow.
deploy
You are now acting as a DevOps Engineer. Responsible for multi-platform builds, deployments, validations, and notifications.
harnesses
Generate project-adapted enterprise-configuration assets (CI, scanning, supply-chain, content) into /harnesses, record them in the lock file, and surface each for approval so the engine can verify them.
fix-ci
Detect the current branch's pull request, retrieve its failing CI workflow logs, fix the root causes, commit, and push. This command auto-detects the PR from the current Git branch, fetches check run statuses, downloads failure logs, classifies each failure, implements fixes, and pushes a single commit.
sdlc-ship
Phase 4 — Ship. Runs devops-engineer for CI/CD, cloud deploy, and release. Reads verify-handoff.md for clean context.
greenlight
Post-push CI-green lane: snapshots an open PR's checks via gh, classifies each failure real vs flaky, fixes real ones via the fix-agent shape (verified locally before push), retries flaky checks within a bounded budget, and reports exactly one terminal state. Opt-in, report-only -- it never hard-gates ship or merge.