DataArt/artisyn-implementation-harness / skills/implementation-harness
implementation-harness
FlaggedInstall the portable AI-dev harness, rebuild its operational graph standalone, harvest project skill improvements into its canonical templates, package direct canonical edits into a portable bundle, or apply canonical updates to an existing project. Use /implementation-harness launch, /implementation-harness reindex-graph, or /implementation-harness patch harvest|prepare|apply.
canonical/skills/dev/TEMPLATE.md
View on GitHub---
name: dev
metadata:
version: 3.2.0
description: >
Full dev lifecycle ({{provider:invoke:dev}} {ticket}): Context Gathering → Ceremony → Plan → Implement
→ Test → Document → Completion, or a single stage via
{{provider:invoke:dev}} context|ceremony|plan|implement|test|document|complete {ticket}.
Trigger: "build {{PROJECT_KEY}}-XX", "implement {{PROJECT_KEY}}-XX", "finish {{PROJECT_KEY}}-XX", or starting a ticket from scratch.
---
The developer wants to run the dev lifecycle for a ticket. Here is their input:
---
## {{provider:input}}
---
## How approvals work
A request to run the dev lifecycle authorises everything routine: reading context,
planning, reversible local edits in scope, running non-guarded checks, updating the
report, moving between stages. That authorisation holds for the whole ticket. Do not
ask again per file, per finding, per step or per phase.
The developer should feel like they are reviewing a senior colleague's work, not
approving every move of a junior one. So:
- **Two stops per phase, both at its end.** The sign-off (`patterns/phase-sign-off.md`):
Context on the written intent, Ceremony on the branch, Plan, Implement, Test and
Document on their results. Then the commit: proposed, run by the developer, checked
in Git before the stage closes. Context is the exception: it has no commit of its
own, its files ride in Ceremony's first commit after the checkout. Completion ends
with one combined ask for external actions.
- **Inside a phase, keep going.** Steps use intelligent checkpoints
(`patterns/approval-intelligent.md`): run the checks yourself, auto-approve when
the content qualifies, and say in one line what you checked. Escalate only on
real doubt.
- **Stop mid-phase only for** something only the developer can do or decide: a
check needing their eyes or their access, a guarded command, a material scope or
architecture choice, accepting a real risk, or an external mutation (tracker,
publish, deploy). Prepare the content first, combine related questions, ask once.
- **Never invent extra gates.** If a stage reference does not name a stop, there is
no stop.
Commits are always the developer's. Each phase ends by proposing the commit(s) and
waiting for them; the next phase starts on a clean worktree.
## Route
See [references/00-router.md](references/00-router.md) for what each stage
produces. Parse `{{provider:input}}` into an optional leading stage word and the
rest:
- `context {input}` → Stage tasks for **Context Gathering** only, then `references/01-context-gathering.md`
- `ceremony {input}` → Stage tasks for **Ceremony** only, then `references/02-ceremony.md`
- `plan {input}` → Stage tasks for **Plan** only, then `references/03-plan.md`
- `implement {input}` → Stage tasks for **Implement** only, then `references/04-implement.md`
- `test {input}` → Stage tasks for **Test** only, then `references/05-test.md`
- `document {input}` → Stage tasks for **Document** only, then `references/06-document.md`
- `complete {input}` → Stage tasks for **Completion** only, then `references/07-completion.md`.
Use this for any story that reached Document but was never archived.
- No stage word (bare ticket or free-form) → run the **full chain**: Entry check,
then Phase 1 through Phase 7 below
- Unrecognised stage word → ask which stage was meant, or whether the full chain
was intended, before doing anything else
**Compatibility:** `/dev-plan`, `/dev-implement`, `/dev-test`, `/dev-document` are
retired — use `{{provider:invoke:dev}} context|ceremony|plan|implement|test|document|complete {ticket}`.
Ticket-tracker activity is entirely `@scrum-master`'s job. Dev only asks it to act
on a ticket by state name ("transition to In Progress", "transition to Done"); the
transition IDs live in `@scrum-master`'s own definition, so the two never drift.
---
## Entry check — every entry, full-chain or direct stage
This is the one place that decides fresh vs. resume. Stage references do not repeat it.
Parse the ticket number from `{{provider:input}}` (`{{PROJECT_KEY}}-\d+`). For free-form
input with no ticket, search existing story folders for a matching local ID
(`DEV-YYYYMMDD-XXXX`) and title; reuse it when found, otherwise generate a
collision-checked one. A local ID does not imply a tracker record.
Then look for existing work:
- Glob `docs/Stories/*{TICKET}*/impl/2-report.md` (use `*{TICKET}*` wildcard — never
`[{TICKET}]*`); include `docs/Stories/archive/`
- If it exists, read the `Status` and `Phase` rows, and check which stage artifacts
exist (`0-intent.md`, `1-plan.md`, archive location)
Decide:
- **No report** → fresh. Full chain starts at Phase 1; a direct stage word starts at
that stage (and Plan will write the missing intent first, see its Step 1).
- **Status `✅ Complete`** → say so and stop. The ticket is finished.
- **Report exists** → one question, same words everywhere:
> "I found existing work for `{TICKET}`: `{phase-row}`. {Artifacts found, one line.}
>
> **resume** where it left off, or **restart** {this stage / from scratch}?"
**Stop and wait.**
- **resume** — full chain: jump to the first stage not ✅ (e.g. `🔨 Implement 🔄`
→ Phase 4; `📝 Document ✅ · 🏁 Completion ⏳` → Phase 7). Direct stage: enter it
and build on what is there; do not redo completed steps or re-ask given sign-offs.
- **restart** — full chain: from Phase 1. Direct stage: that stage from its first
step, overwriting its own artifact only (`0-intent.md` for Context, `1-plan.md`
for Plan); earlier stages' artifacts stay.
The story folder found here is `$IMPL_DIR` (`docs/Stories/[{TICKET}] {name}/impl/`,
or its `archive/` twin) for every stage and every metrics command from now on. Stages
do not search for it again; a direct folder path given as input is used as is.
Run metrics `recover --dir "$IMPL_DIR"` when a `stat.json` exists, before recording
anything. Then apply **Stage tasks** below and go to the target phase.
---
## Stage tasks — apply once, every entry (mandatory)
Do this on every entry, full-chain or single-stage, resumed or fresh, before the
target stage's first metrics call. It is a step, not a reminder-driven aside: do
not wait for a system reminder about progress tools, and never finish a stage with
its native task still open.
Apply `{{provider:skills_root}}/dev/patterns/task-tracking.md`. It inspects existing
state (the native tracker or `2-report.md`) and decides create vs. reuse vs.
reconcile; this router does not. For the full chain the unit list is the seven stages
(Context Gathering, Ceremony, Plan, Implement, Test, Document, Completion) in order.
For a direct single-stage call it is that one stage. Mark a stage `in_progress` on
entry and `completed` only when the phase banner shows it done. A stage reference
tracks its own child units (plan steps, in Implement's case).
---
## Metrics
Apply `patterns/metrics.md` throughout. Each ticket has `impl/stat.json` beside
`2-report.md`. Use `scripts/timing.mjs` with an explicit `--dir` for every metric
change; never edit the JSON by hand. Each stage reference says where its `stage
start`, `wait start`, `resume` and `stage complete` calls go; the phase sign-off
pattern owns the sequence around every stop. The entry check runs `recover` on resume.
Keep unknown measurements explicit.
---
## Phase tracker
Show this banner at every phase transition, with the active phase marked:
```
🎯 Context [status] · 🚀 Ceremony [status] · 📋 Plan [status] · 🔨 Implement [status] · 🧪 Test [status] · 📝 Document [status] · 🏁 Completion [status]
```
`[status]` is `🔄` (active) · `✅` (done) · `⏳` (pending). Example when entering Plan:
```
🎯 Context ✅ · 🚀 Ceremony ✅ · 📋 Plan 🔄 · 🔨 Implement ⏳ · 🧪 Test ⏳ · 📝 Document ⏳ · 🏁 Completion ⏳
```
---
## Phase 1 — Context Gathering
Follow `references/01-context-gathering.md`. Ends with a phase sign-off on the
intent. Announce the outcome and continue; direct-stage entry ends here.
## Phase 2 — Ceremony
Follow `references/02-ceremony.md`. Ends with a phase sign-off on the branch.
Announce the outcome and continue; direct-stage entry ends here.
## Phase 3 — Plan
Follow `references/03-plan.md`. Ends with a phase sign-off. Announce the outcome
and continue; direct-stage entry ends here.
## Phase 4 — Implement
Follow `references/04-implement.md`. Ends with a phase sign-off. Announce the
outcome and continue; direct-stage entry ends here.
## Phase 5 — Test
Follow `references/05-test.md`. Ends with a phase sign-off. Announce the outcome
and continue; direct-stage entry ends here.
## Phase 6 — Document
Follow `references/06-document.md`. Ends with a phase sign-off. Announce the
outcome and continue; direct-stage entry ends here.
## Phase 7 — Completion
Follow `references/07-completion.md`. It archives the story, finalises metrics,
hands over the commit and PR text, offers the tracker and tech-debt actions once,
and ends with a clear "this ticket is complete" message. There is nothing after it.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |