prekuter/dryforge / agent-plugin/skills/go
go
VerifiedCarry out the intent approved in `ready`, as meant, and prove it with checks that actually ran. Comes back to you if the work would change what you approved, keeps the project docs in step, and never merges on its own. Use when the user invokes the `go` skill after `ready`. Requires git.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 28.4 KB |
| references/foundation-format.md | markdown | 5.3 KB |
| references/graph-contract.md | markdown | 4.4 KB |
| references/harness-format.md | markdown | 28.8 KB |
| references/harness-lifecycle.md | markdown | 6.6 KB |
| references/harness-review.md | markdown | 7.3 KB |
| references/implementer-prompt.md | markdown | 5.7 KB |
| references/orchestration.md | markdown | 24.8 KB |
| references/reviewer-prompt.md | markdown | 3.4 KB |
| references/spec-review-prompt.md | markdown | 2.5 KB |
references/implementer-prompt.md
View on GitHub# implementer-prompt.md — the implementer subagent prompt
One subagent implements one task, **verified-first** (test-first where it fits), in its pinned
worktree, then commits and reports. The prompt has **required elements** (mandated — the scaffold)
and **wording you adapt** (the example below is one phrasing, not a fixed script).
## Required elements (every implementer prompt must pin)
- **Task contract** — the task's goal and **work targets (files | state | external resource)**,
from the plan (behavioral, not premature code). **Defer plan-owned decisions to the plan — do not
restate them.** Names (symbols, files, endpoints), shared conventions, and interfaces the plan
already fixes must be referenced ("follow the plan's naming"), not re-specified in the prompt: a
prompt that restates them in different words makes parallel implementers diverge. Pin only what
*this* task adds. For a task that declares **no file diff** (a state-change / operational /
external-config target), verification is by **commit message + captured external evidence**
(command exit / render / API or state response), not necessarily a file diff — the
captured-evidence floor still holds.
- **Spec section** — the spec requirement this task realizes, **quoted inline in this prompt**
(the task worktree has no `.dryforge/` files to read — `.dryforge/` is gitignored). Build to the
spec, not just to the task line ("correct" = matches the spec).
- **Hard gates** — the relevant non-negotiable constraints from the handoff.
- **Verify-first, right-sized** — drive the work against the task's **real verification gate**: the
project's verify commands, discovered by the producer (which may be a typecheck/lint/test set, a
build/run, or a single command — not every stack has all three; never assume the triad).
**Consume the producer-set tier — don't re-derive "is this risky?" from scratch.** This task is
classified RISK=<tier> by the producer. RISKY → test-first RED→GREEN against the project's verify
commands. MECHANICAL → a confirming check against the gate, no RED ceremony. NONE → appropriate
evidence (build / validate / render / apply), no unit-test ceremony. (Test-first means: failing
test → confirm **RED** → minimal implementation → confirm **GREEN** → refactor.) If the tier
looks wrong for what you find, return `DONE_WITH_CONCERNS` and say so — do not silently upgrade or
skip. If the producer omitted the tier, judge risk yourself at build time. The floor is
*captured-evidence verification*, not ceremony — **but testable behavior with real risk left
untested is still needs-fix** (the tier sizes the ceremony, it is not an excuse to skip tests that
matter).
- **Shared-write constraints** — which files are this task's to write, and which **not** to
touch (the shared/registration files a later wiring step owns).
- **Worktree pin** — the absolute worktree path + branch; **omit isolation** (do *not* enable the
platform's worktree-isolation dispatch option — the orchestrator already
created the worktree; see `orchestration.md`); verify with `git rev-parse --show-toplevel` before
editing.
- **Report contract** — commit when done; return the structured summary below. Do not dump diffs.
## Example skeleton (one wording — adapt; the elements above are required)
```
Implement <task id / goal> in the worktree at <ABS PATH> (branch <name>).
First: `git rev-parse --show-toplevel` must equal <ABS PATH> — if not, stop and report.
Spec to satisfy: <spec slice>. Hard gates: <relevant gates>.
Files yours to write: <...>. Do NOT touch: <shared / registration files>.
Verify against the task's gate (the project's verify commands). This task is RISK=<tier>: RISKY →
test-first (failing test → RED → minimal impl → GREEN → refactor); MECHANICAL → confirming check, no
RED; NONE → appropriate evidence (build / validate / render / apply). If the tier looks wrong, return
DONE_WITH_CONCERNS. For a no-file-diff target, evidence is commit message + captured external
response. Run the project's verify commands; capture exit codes.
When done: commit, then return ONLY the structured summary. Do not inline diffs.
```
## Structured return
- `status`: `DONE` | `DONE_WITH_CONCERNS` | `NEEDS_CONTEXT` | `BLOCKED`
- `files_changed`: paths — or, for a declared no-file-diff task (state-change / operational /
external-config), the work target touched plus the commit that records it
- `verification`: the **literal command(s) run and their exit codes** (evidence, not "passed") —
for a testable task, the tests added + their run; for a non-unit-testable deliverable, the gate
actually run (build / validate / render / ...). A claimed pass with no command/exit code is
needs-fix; and **omitting tests on a task that has real testable behavior is needs-fix** (the
no-test path is only for genuinely non-unit-testable or trivial work)
- `concerns`: anything the orchestrator should weigh (or what is blocking)
## escalate-don't-guess (for the implementer)
If the task is ambiguous, conflicts with the spec, or can't be done as written — **do not
guess or quietly improvise.** Return `NEEDS_CONTEXT` or `BLOCKED` with specifics; the
orchestrator decides and escalates to the user if needed. A wrong guess is worse than a stop.
You are in a fresh session with no live user conversation — do NOT ask the user directly. Return
`NEEDS_CONTEXT` or `BLOCKED` instead; the orchestrator relays escalations to the user.
**Honor any shared-resource expectation the task declares** (a fixture, a shared file/state, an
external resource the plan says should be in a given state). If you find that shared resource in an
unexpected state, return `BLOCKED` with the specifics — do not work around it silently.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file). | Stars: 198k | Updated | Claude CodeCodex | Unverified | Install |
| doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |