sdlc
Coordinate Epoch feature delivery end to end with persona-first BDD/TDD, stacked PRs, incremental commits, automatic reviews, anti-slop + DESIGN.md gates, Pact at integration boundaries, evidence packs posted on PRs, explicit subagent/branch/worktree lifecycle, and repo hygiene (anti-bloat, cohesion/coupling, cleanup). Evolve agent skills with Microsoft SkillOpt / SkillOpt-Sleep and harness-aware workflows. Keep documentation accurate against the freshness policy and docs standards. Invoke as sdlc / /sdlc with subcommands: help, loop (default), finish, clean, review, plan, brainstorm, dispatch, test, evidence, gate, eval, skills, docs, init. Use when the user says sdlc, sdlc help, finish the session, clean branches/worktrees, improve agent skills or docs, or wants the full development loop.
Informations de source
- Dépôt
- Tyler-R-Kendrick/epoch
- Dernière activité de la source
- 20 août 2026 à 19:26
- Langue détectée de SKILL.md
- anglais
- Étoiles
- 0
- Forks
- 0
Options d'installation
Le prompt qui vérifie d'abord la source est sélectionné par défaut. Vous pouvez passer à une commande directe ou télécharger une copie locale.
Vérifiez les fichiers source
Lisez SKILL.md et les fichiers associés affichés par SkillsMP avant de décider de l'installer.
Explorateur de fichiers
23 fichiersAffichage de SKILL.md
SKILL.md
Instructions source · Aperçu en lecture seule- name
- sdlc
- description
- Coordinate Epoch feature delivery end to end with persona-first BDD/TDD, stacked PRs, incremental commits, automatic reviews, anti-slop + DESIGN.md gates, Pact at integration boundaries, evidence packs posted on PRs, explicit subagent/branch/worktree lifecycle, and repo hygiene (anti-bloat, cohesion/coupling, cleanup). Evolve agent skills with Microsoft SkillOpt / SkillOpt-Sleep and harness-aware workflows. Keep documentation accurate against the freshness policy and docs standards. Invoke as sdlc / /sdlc with subcommands: help, loop (default), finish, clean, review, plan, brainstorm, dispatch, test, evidence, gate, eval, skills, docs, init. Use when the user says sdlc, sdlc help, finish the session, clean branches/worktrees, improve agent skills or docs, or wants the full development loop.
# SDLC coordinator
You are the **parent/coordinator** for product work in this repository. Own the user dialogue,
plan, shared files, **stack topology**, reviews between layers, and merge gate. Delegate
implementation to subagents (one issue/layer each) whenever possible.
Everything starts from **documented personas** (humans and agents-as-users). Specs, designs,
tests, and code exist only when they change an intended persona outcome. Prefer the minimum
necessary design/spec to produce that outcome.
## First moves
1. Parse the subcommand (table below). If `help` / `--help` / `-h`, load
[references/help.md](references/help.md), print usage, and **stop** (unless they named a
specific command to explain).
2. Otherwise load **only** the matching reference.
3. Run `sdlc init` probes when mirrors, `.sdlc/`, or `gh stack` may be missing.
4. Read durable state: [`.sdlc/state/current.yaml`](../../.sdlc/state/current.yaml) and
`docs/plans/<initiative>/sdlc-state.md`. Reconcile before new work
(`references/dispatch.md` § Resume).
## Subcommands
| Command | Reference | Purpose |
|---|---|---|
| `help` / `--help` / `-h` | [references/help.md](references/help.md) | Print usage, commands, flags; optional `help <command>` |
| `(default)` / `loop` | this file + phase refs | Full persona-first loop → stack closeout |
| `finish` / `--finish` | [references/finish.md](references/finish.md) | Commit, rebase, PR, bottom-up review, squash-merge session PRs |
| `clean` | [references/stages/clean.md](references/stages/clean.md) | Delete merged session branches/worktrees (local/remote flags) |
| `review` | [references/stages/review.md](references/stages/review.md) | Security / design / architecture review between PRs |
| `brainstorm` | [references/brainstorm.md](references/brainstorm.md) | Adversarial + rubber-duck idea hardening |
| `plan` | [references/planning.md](references/planning.md) | Plan mode + stack layer slicing |
| `dispatch` | [references/dispatch.md](references/dispatch.md) | Subagent dispatch + handback |
| `test` | [references/stages/test.md](references/stages/test.md) | Persona Gherkin + Playwright + Pact (no new full-stack e2e by default) |
| `evidence` | [references/stages/evidence.md](references/stages/evidence.md) | Publish traces/video/snapshots + NL summary |
| `gate` | [references/stages/gate.md](references/stages/gate.md) | `gate:commit` / optional `gate:push` / `verify` |
| `eval` | [references/stages/eval.md](references/stages/eval.md) | Rubric scoring for self-improving loops |
| `skills` | [references/skill-evolution.md](references/skill-evolution.md) | SkillOpt / Sleep; promote chat patterns; harness workflows |
| `docs` | [references/documentation.md](references/documentation.md) | Freshness matrix, accuracy audit, ADR/docs standards |
| `init` | [references/stages/init.md](references/stages/init.md) | Mirror skill symlinks, ensure `.sdlc/`, probe `gh stack` |
### Flag cheatsheet
```text
sdlc help [command]
sdlc finish
sdlc clean [--local] [--remote] [--worktrees] [--merged-only] [--dry-run] [--force]
sdlc review [--security] [--design] [--architecture] # default: all three
sdlc gate [--push] [--verify]
sdlc test [--persona <tag>] [--feature <path>]
sdlc evidence --feature <slug>
sdlc eval --initiative <slug>
sdlc skills [--promote] [--sleep] [--opt] [--workflow] [--dry-run]
sdlc docs [--audit] [--fix] [--matrix] [--adr] [--dry-run]
sdlc init [--check]
```
`finish` is **explicit authorization** to push, open/update PRs, and squash-merge session work
(billing-red CI does not block merge). Safety still applies: no secrets, no force-push, no
deleting unrelated user work.
## Hard rules (always on)
1. **Incremental commits** after each red→green unit. Never `SKIP_GIT_HOOKS` / `--no-verify`.
2. **Stacked PRs** for 2+ dependent layers via non-interactive `gh stack`
([references/stacked-prs.md](references/stacked-prs.md)).
3. **`sdlc review` between stacked PRs** before bottom-up squash-merge.
4. **Persona minimum** — [references/persona-minimum.md](references/persona-minimum.md).
5. **Tests only** for components exercised by `@persona.*` scenarios in `features/*.feature`.
6. **New work:** Gherkin + Playwright driver + **Pact** at integration boundaries. Do **not**
add new full-stack e2e when Pact covers the boundary (existing e2e stays until migrated).
7. **Gates:** `npm run gate:commit` before every commit (anti-slop `lint:oxlint`, `design:lint`,
`design:audit`, Community Web a11y + design chrome lint). See [docs/anti-slop.md](../../docs/anti-slop.md)
and root [DESIGN.md](../../DESIGN.md).
8. **Decisions** persist under [`.sdlc/decisions/`](../../.sdlc/decisions/) (machine YAML). Material
choices also get an ADR under `docs/design-decisions/`.
9. **Evidence** for completed features: [references/stages/evidence.md](references/stages/evidence.md).
Publish the pack under `docs/evidence/<slug>/`, then post a visible `## SDLC evidence` section
on every related PR (**body + sticky comment**) that links that pack path before squash-merge.
Saying “evidence” without the `docs/evidence/…` path and PR block is incomplete.
10. **Operations lifecycle** — create/destroy subagents, branches, and worktrees only per
[references/operations.md](references/operations.md). Parent owns topology and cleanup.
11. **Repo hygiene** — no file/folder explosion; high cohesion / low coupling; clean merged
worktrees and never commit caches. See [references/repo-hygiene.md](references/repo-hygiene.md).
12. **Skill evolution** — prefer Microsoft SkillOpt / SkillOpt-Sleep over lean forks; promote
repeated chat/coding patterns into skills; emit harness workflows only for the active
host ([references/skill-evolution.md](references/skill-evolution.md)).
13. **Documentation** — same-PR freshness, no orphans, accurate claims vs code/features;
`npm run docs:check` before finish ([references/documentation.md](references/documentation.md)).
## Default loop (when bare `sdlc` / `loop`)
1. `brainstorm` → survive adversarial + rubber-duck passes.
2. `plan` → draft cascade; slice stack layers; record decisions in `.sdlc/decisions/`. Prefer
extending existing packages/modules over new directories (`repo-hygiene.md`).
3. Issue capture (`references/linear-planning.md`) — one issue per layer when possible.
4. Create stack branches / local worktrees only as needed (`operations.md`); then `dispatch`
— implementers use `test` (red) → implement → `gate` → commit; apply `sdlc docs --fix`
for touched surfaces before handback.
5. After each layer PR: `review` (default all facets, including docs accuracy) → `evidence`
(PR sticky comment) → repair → only then next layer.
6. Close stack bottom-up (`stacked-prs.md`); refresh evidence + `eval` before marking Done.
7. Stop agents; `clean --merged-only` (+ `--worktrees` / `--remote`); drop empty dirs / dead
paths that the initiative made obsolete.
8. If the same agent mistake or user correction recurred this initiative, run
`sdlc skills --promote` (and `--sleep` when transcript optimization is warranted).
9. If docs drifted or users hit the same doc gap twice, run `sdlc docs --audit` / `--fix`.
## Role split
| Actor | Owns | Does not own |
|---|---|---|
| **Parent (this skill)** | User dialogue, plan, Linear, shared files, stack topology, reviews, evidence-on-PR, merge, clean | Implementing every layer when a subagent is available |
| **Subagent** | One layer, red/green, incremental commits, local `gate:commit`, drafting evidence pack files | Linear, stack restructure, merging PRs, deleting branches/worktrees |
| **Reviewer** | Independent `sdlc review` + checklist (read-only) | Edits, merges |
Lifecycle detail (when to spawn/stop agents, create/destroy branches and worktrees):
[references/operations.md](references/operations.md). Anti-bloat and cleanup cadence:
[references/repo-hygiene.md](references/repo-hygiene.md). Skill / workflow evolution:
[references/skill-evolution.md](references/skill-evolution.md). Documentation accuracy:
[references/documentation.md](references/documentation.md).
## Subagent incremental checkins
1. After each red→green unit, stage deliberately and commit.
2. Run `npm run gate:commit` (hooks already do this) — never bypass.
3. Conventional, scoped commit messages (why if non-obvious).
4. Multiple commits per layer are expected; squash only at **merge** time.
5. Do not create extra branches/worktrees beyond the assigned layer; do not run `sdlc clean`.
6. Prefer extending existing modules; justify any new directory/package in the handback.
7. Never tell implementers to run `pnpm agent:check` — Epoch uses `npm run gate:commit`.
## Install / probe
```bash
npm run skills:mirror-sdlc # or: sdlc init (also installs SkillOpt once)
gh extension install github/gh-stack
git config --local rerere.enabled true
git config --local remote.pushDefault origin
```
Host trees (`.agents/skills`, `.claude/skills`, `.grok/skills`, `.cursor/skills`) **symlink** to
tracked `skills/sdlc`. Do not maintain duplicate copies.
## Reference index
See [references/index.md](references/index.md) for the full progressive-disclosure map.
## Invocation examples
```text
/sdlc help
/sdlc help finish
/sdlc
/sdlc finish
/sdlc clean --merged-only --dry-run
/sdlc review --design --security
/sdlc test --persona @persona.github_open_source_contributor
/sdlc evidence --feature community-web-receipts
/sdlc gate --push
/sdlc eval --initiative anti-slop-zero
/sdlc skills --promote --dry-run
/sdlc skills --sleep
/sdlc skills --workflow
/sdlc docs --fix
/sdlc docs --audit --dry-run
/sdlc init
```
Voir sur GitHub