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/harness-review.md
View on GitHub# harness-review.md — verifying a generated/updated harness (force-load)
The verification criteria for a project harness. Used by **migration's REVIEW phase** and by
**`go`'s final-review lenses 3–4** — one source of truth, so both judge the harness the same way.
The spec being verified against is `harness-format.md`.
This is an **adversarial pass**: the reviewer's job is to find holes, not to bless the work
(self-judgment is the weak move). The four dimensions below are *what to
check*, applied with judgment — not a box-ticking checklist that earns a pass by being walked.
## Dimension 1 — content (does each file substantively meet its spec?)
For each file, check it against `harness-format.md`'s spec for that file, then against the
cross-cutting quality bar:
- **Meets its own spec** — has the *kinds* of information harness-format requires for that file, at
that file's altitude, and clears that file's quality floor (e.g. business-rules: every rule
verifiable, each "must" paired with a "must not," edges concretely disposed; security: failure
paths named, explicit "not allowed" stated; standards: every rule has a violation criterion).
- **Five principles** — non-derivability (nothing the code already reveals), work-changing (would
change the next agent's work), density (every sentence carries a fact), project-specificity (no
universal truths), consequence-of-absence (removing it would break something).
- **Four techniques** — verifiable statements, counter-cases stated, mechanism described (not just
outcome), edge cases concrete (no "handled appropriately").
- **Language** — the harness is written natively in the **language the user communicates in**
(language-agnostic; the specific language is discovered, not assumed). A harness that doesn't match
the user's language is a defect. (The reviewer is told the user's language by the orchestrator at
dispatch — in both migration and go the finished work is verified by an independent dispatched
reviewer, never self-reviewed.)
- **Harness describes the project only (blocking)** — flag any file that describes itself (its state,
thinness, lifecycle, or *its own purpose / how to read it* — e.g. "read this instead of the code"),
its origin (that it was generated / migrated / is a first-cycle artifact; status.md listing the
documentation layer itself as a delivered capability), or that uses the **generating tool's name or
dryforge's coined vocabulary to label the doc system** ("dryforge" anywhere; calling the docs "the
harness"/"하네스", a section "Harness navigation", or naming "3-doc"/"delta"/"wave"). Section titles
use the project's plain terms. It describes the project, not the document. ("dryforge" is a reliable
sweep; for "harness" use judgment — a project's own *test* harness is legitimate, labeling the doc
layer a harness is not.)
- **Entry-point navigation is one whole-harness tree** — CLAUDE.md/AGENTS.md show the navigation as a
single project-root directory tree (`├──`/`└──`) covering the entry files + `docs/` + module
`AGENTS.md` at their dirs; flag a table, a flat list, or a docs-only tree with modules tacked on.
- **No baked scope-freeze in any doc** — flag *any* doc that writes planned-but-unbuilt scope (a
status.md "remaining" item) as a permanent constraint: a hard gate or standards rule saying "don't
build X yet", a business-rule calling a planned capability "impossible by construction", or a "keep
shape Y" forward-compat gate where a remaining item will change shape Y. Planned scope is a *status*,
not a permanent invariant — it falsely conflicts with the cycle that builds it. Describe the current
state, not a permanent ban; hard gates and domain laws are permanent only.
The failure mode to hunt: a **hollow shell** — structurally present, substantively empty. A file
that reads as generic boilerplate, restates what code shows, or fills a section with modifiers
("modern," "robust," "appropriate") fails this dimension even though the section "exists."
## Dimension 2 — format (self-containment)
- **No cross-references between `docs/` files** — no "see architecture.md"; each file states what it
needs at its own altitude.
- **Each file at one altitude** — no upper-altitude detail replicated below, no lower-altitude
implementation mentioned above. The same topic appearing in two files is fine *only* if each
treats it at its own altitude.
- **No 3-doc references** — nothing points at `.dryforge/` (the 3-doc is archived after the task;
the pointer would dangle).
- **No constraint-ID labels (blocking).** Flag any constraint-ID carried into the harness — the
3-doc's `INV-N` / `T-N`, or any `XXX-N` scheme — and any reference to another doc's invariant/rule
*by label*. These dangle once the 3-doc is archived and violate self-containment; invariants must be
stated **by content**. A sweep for `\b[A-Z]{2,}-[0-9]+\b` across the harness catches most.
- **Module AGENTS.md** references no other module AGENTS.md and no `docs/` file.
## Dimension 3 — completeness (required files present)
- All `docs/` files exist (the 7 core + tracking: status.md, decisions/index.md, findings.md).
- CLAUDE.md and AGENTS.md both exist, with **identical content**.
- An AGENTS.md exists for every identified module (module-identification heuristic in
harness-format.md). A missing module AGENTS.md for a real module is a defect.
- An empty file is correct only when its category genuinely doesn't apply; an empty file for an
applicable category is a defect (hollow shell), caught by Dimension 1.
## Dimension 4 — against the source (omission vs hallucination)
Cross-check the harness against the code, **both directions**:
- **Omission** — an intent / constraint / mechanism that the code **cannot derive on its own** is
missing from the docs. This is the real omission. *Code-derivable facts being absent is not an
omission* — by the non-derivability principle, the harness deliberately omits them.
- **Hallucination** — the docs assert something the code does not contain. Flag it — **except**
content that legitimately derives from **future scope**: status.md's "remaining" items,
business-rules' not-yet-implemented rules, and other forward-looking content sourced from the
Foundation/design. For those, code absence is *normal*, not a hallucination. Distinguish "the doc
describes a planned future state" (correct) from "the doc describes a present state the code
contradicts" (a real hallucination).
## Findings → disposition
Each finding is one of:
- **Internally resolvable** (a hollow section you can deepen from sources already in hand, a
cross-reference to inline, an altitude violation to move) → fix it directly.
- **Needs user intent** (a gap only the user can fill — a domain rule, a policy decision, a
rationale not on the record) → do not invent it; raise it to the user (migration: ask in the user
gate; go: surface as a blocking finding the orchestrator escalates).
In `go`'s final review, harness findings carry the same blocking/advisory split as code findings.
## Universality guard
Stack-agnostic. The criteria judge information *kind*, density, and altitude — never conformance to
a particular stack. What counts as a module, a contract, or a representative flow is whatever the
project actually is, discovered at runtime.
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: 472k | 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 |