Use when asked to score a repo's delivery posture, audit CI/CD, check branching/environments/IaC/flags/rollback readiness, or adopt/scaffold the pipeline standard. Runs a zero-dependency Node checker (39 rules across descriptor & declared posture, branching & PR workflow, CI/CD stages, environment tiers, infrastructure as code, feature flags, rollback & recovery, hotfix & incident flow, and the sign-off ledger), reads the scorecard, and helps fix or scaffold what's missing — calibrated to declared failure cost (solo/team/critical).
Instrucciones de origen · Vista previa de solo lectura
name
pipeline
description
Use when asked to score a repo's delivery posture, audit CI/CD, check branching/environments/IaC/flags/rollback readiness, or adopt/scaffold the pipeline standard. Runs a zero-dependency Node checker (39 rules across descriptor & declared posture, branching & PR workflow, CI/CD stages, environment tiers, infrastructure as code, feature flags, rollback & recovery, hotfix & incident flow, and the sign-off ledger), reads the scorecard, and helps fix or scaffold what's missing — calibrated to declared failure cost (solo/team/critical).
A testable delivery-operations standard: 39 rules, each backed by a check a zero-dependency Node runner executes on a repo at rest. Blockers fail CI (exit 1); the judgment calls a script can't make resolve via a dated sign-off ledger. The throughline: complexity scales with failure cost — you declare your tier (solo / team / critical) once, and the standard holds you to exactly that much process, no more, no less. A delivery-practices doc drifts; this is the delivery checklist as an exit code.
Runs natively under Hermes and Claude Code (and any agent that loads SKILL.md). The runner is portable — plain Node, no agent-specific dependency. Sibling of baseline-skill (repo readiness), not a fork: baseline answers "ready to build and maintain?", pipeline answers "can it ship, recover, and prove it?" — installable alongside, sharing one records/judgments/ ledger format.
When to Use
"score this repo's delivery posture", "run pipeline", "audit the CI/CD setup"
"are our environments / deploys / rollback actually in order?"
"set up / adopt / scaffold the pipeline standard here"
"fix the pipeline failures", "get delivery to green"
"what does ENV-04 check", "why did IAC-04 fail"
Don't use for: build/test/secrets/reproducibility readiness (use baseline), general logic/bug review (use a code-review skill), or runtime/cloud truth — pipeline checks delivery mechanisms and declarations at rest; it can never see your actual infra, staging data, or flag states. What a script can't see resolves via the sign-off ledger.
Setup — resolve the toolkit path first
$SKILL_DIR = the absolute path of the directory containing this file. Resolve it to a concrete path before running anything (never pass the literal $SKILL_DIR to the shell). Typical locations:
The unified CLI is pipeline.mjs — node "<abs>/pipeline.mjs" <command> (check, jdg, help); pipeline check … delegates to , still the checker. Both load the rule set ( manifest + modules) and from their own directory, so always invoke ; don't copy them away from the rule set + . Requires on PATH — if is missing, say so rather than guessing. : forge-reading rules degrade to honest, labeled SKIPs offline or token-denied.
check.mjs
rules.json
rules/
src/
by absolute path
src/
Node ≥ 18 and git
node
gh is optional
Co-located files: pipeline.mjs (CLI entry: check / jdg), check.mjs (the checker), rules.json + rules/ (the rule-set manifest + the 39 rules, one module per family), schema/ (descriptor + judgment + flag-registry schemas), config.example.json, templates/ (scaffolds), config-presets/ (per-tier descriptor presets), REFERENCE.md (full reference), GLOSSARY.md (term definitions).
The descriptor — declare your failure cost first
Everything hangs off pipeline.repo.json at the repo root: type, profile (solo|team|critical), branching_model, default_branch, environments[], iac, uses_feature_flags, branches (optional tier-branch names; unset keys fall back to the branching_model conventions), recovery_objective. The profile field is the philosophy mechanized:
solo — the pipeline must exist, be automated, be alive; rollback documented. No required review, no staging, no IaC mandate.
team (adds) — protected default branch, PR review ≥1, declared environments each with an automated deploy path, IaC not console-clicking, documented hotfix path.
Tiers are cumulative; off-tier rules SKIP as n/a and never count against you. Declare more than you carry (or carry more than you declare) and DESC-03 (the inverse-nag) says so — resolve it in the descriptor, in the repo, or as a dated deviation judgment, never silently.
Judgments — the sign-off ledger
The rules a script can't decide (staging data realism, rollback drills, reviewed recovery objective) and the exceptions you consciously accept (a manual deploy, an external CI system, a kept expired flag) are ledger records, not chat:
One judgment, one file: records/judgments/JDG-NNNN.json. Every judgment is dated, owned, scoped, reasoned — and expires (review_by); a tripwire voids it when the accepted world changes. A kind: sign-off judgment whose subject is a manual rule's id satisfies that rule while unexpired; GOV-01 makes ledger health itself a scored blocker, so an expired judgment can't silently void the rule it was satisfying. A break-glass record names the gate it relieved (HOT-04) and lands via its own PR. Never fake a sign-off: a real dated judgment by the user, or nothing.
Modes
Figure out intent from the user's words; default to score.
score (default)
Pick the target repo: an explicit path, else the current working directory. Confirm it looks like a repo (a manifest or .git).
Run the runner:
node "$SKILL_DIR/check.mjs" --repo <target>
--json — machine output instead of the scorecard.
--no-exec — reserved; v1 has no exec-class checks (kept for CLI parity).
Completion criterion: you have the readiness %, the blocker count, and each FAIL/notable WARN with its one-line detail.
Present it: lead with blockers (they fail CI), then warnings worth fixing, grouped by family. Don't dump all 39 rows — summarize and offer to fix or scaffold. Say which rules SKIPs affected (off-tier, opt-in, forge-offline) — a critical repo scored offline never had its prod gate verified; say that out loud.
init — "set up / adopt / scaffold pipeline"
Descriptor-first, always. The repo's pipeline.repo.json is written before anything else — it's the one file pipeline requires (schema: schema/repo.schema.json), and every profile/cross-check derivation reads it. Its type supersedes filesystem auto-detection.
Write the descriptor. Copy the closest tier preset (or the blank template), then set the fields to reality:
cp"$SKILL_DIR/config-presets/team.repo.json" <repo>/pipeline.repo.json # or solo / critical; templates/pipeline.repo.json is blank
Choose the tier by honest failure cost (what breaks, for whom, if this ships broken?) — not by aspiration. DESC-03 will hold the declaration to the observed reality.
Tune the checks (optional). Copy config.example.json to <repo>/pipeline.config.json and set thresholds (stale_branch_days, rollback_drill_days, prod_environment_name, uses_feature_flags, required_check_names). Opt-in *_globs keys stay empty until adopted.
Scaffold only what's missing (never overwrite without asking):
mkdir -p <repo>/records/flags && cp"$SKILL_DIR/templates/flag.json" <repo>/records/flags/FLAG-0001.json # only if flags are used — opt-in familycp"$SKILL_DIR/templates/runbook-rollback.md" <repo>/RUNBOOK.md # RB-01, if no rollback section exists
Wire the pipeline job into CI as a required check (snippet in REFERENCE.md) — the standard can't rot once it's enforced on every PR.
Run a first score — DESC-01 confirms the descriptor is present, DESC-02 that it schema-validates.
Completion criterion:pipeline.repo.json exists and validates (DESC-01 + DESC-02 PASS), node check.mjs --repo <repo> runs, and every scaffolded artifact is accounted for.
fix — "get this to green"
Score first. For each blocker/warn to address, apply the rule's own fix field (read it from rules.json) as concrete edits — add the missing rollback section, protect the default branch, git-ignore the tfvars, add the drift cron, declare the environments.
For manual (sign-off) rules, don't fake the check — do the judgment with the user (staging data realism, the rollback drill, the recovery-objective review) and record it dated and expiring: pipeline jdg new --kind sign-off --subject <RULE-ID> --reason "..." --review-by <date>.
For conscious exceptions (manual deploy by design, external CI, a justified long-lived branch), record deviation / risk-acceptance judgments the same way — named, reasoned, expiring.
Re-score to confirm.
Completion criterion: re-score shows the targeted rules resolved and no new blockers introduced.
explain — "what does ENV-04 check", "why did IAC-04 fail"
Read the rule from rules.json (title, rationale, fix, source, check) and explain it plainly plus what the runner actually looked for — including which plane answered (tree / history / forge / ledger) and whether a forge SKIP means "fine" or "unknown". For unfamiliar jargon (sign-off ledger, break-glass, drift, FDRT, inverse-nag, …) point to GLOSSARY.md.
Rule-set integrity — --self-check
The rule set validates itself:
node "$SKILL_DIR/check.mjs" --self-check
Exits 1 on any rule with a missing/typo'd applies_to, an unknown check-kind / profile / severity / category / requires key, a duplicate id, an orphan type/profile, a non-deterministic blocker (including any severity_by_profile escalation), or a non-judgment sign-off — and prints a per-type coverage matrix. Use it if you edit the rule modules under rules/ (or the rules.json manifest), and wire it into CI so a malformed rule set can't merge.
How the runner decides (so you can read detail lines)
PASS / FAIL / WARN / SIGN-OFF / SKIP per rule; only a blocker FAIL sets exit 1.
SKIP = the rule didn't apply: applies_to excludes the repo's type, an off profile tier (profile 'critical' off), the FLAG opt-in is inactive, or nothing to check (incl. forge unreachable). A skip never counts against readiness — but forge SKIPs on gating rules (BR-03, ENV-04) mean unverified, not fine; report them as such.
Profiles come from the descriptor's declared tier: solo = core only; team = core+team; critical = everything. Two rules escalate severity by tier (BR-03, ENV-04: warn at team, blocker at critical).
Certainty is labeled: deterministic (presence/absence, set comparison), heuristic (line-scan inference — never dressed up as proof), judgment (ledger only).
The runner is zero-dependency and crash-resilient: an unevaluable check degrades to SKIP, never crashing the run.
Common Pitfalls
Copying check.mjs away from the rule set (rules.json + rules/) + src/. It loads them from its own directory — invoke by absolute path instead.
Presenting a warn as a blocker (or vice-versa). Severity is in the rule modules and the runner output — never upgrade/downgrade it. Two rules legitimately change severity by declared tier (BR-03, ENV-04) — that's the descriptor talking, not you.
Faking a sign-off. Manual rules exist because a script can't judge them; record a real dated judgment (pipeline jdg new --kind sign-off), don't rubber-stamp.
Declaring critical for the badge. The tier is a commitment: critical activates the sign-off surface (drills, staging realism, reviewed objectives) and escalates two blockers. Declare the tier whose process you'll actually carry — DESC-03 compares the declaration to reality.
Reading an offline score as a green score. Forge rules SKIP labeled when gh is unreachable; a critical-tier repo whose prod gate was never verified is not "passing" — say so.
Gaming a warn to hit 100%. An honest advisory warn beats presence-theater. 0 blockers is the bar; the ledger exists for everything else.
Verification Checklist
Ran the runner by its absolute path with --repo <target>
Reported blockers first, then warnings, grouped by family (not a 39-row dump)
Named which rules SKIPs affected (off-tier / opt-in / forge-offline) — and said "unverified", not "fine", for gating rules that skipped
For fix: re-scored and confirmed no new blockers
For init: picked a tier preset by honest failure cost, scaffolded only what was missing, ran a first score
Any sign-off is a real dated judgment, not a rubber stamp
--self-check still passes if the rule set (rules.json / rules/*.json) was edited
