ワンクリックで
auto-frame
Bound and de-risk a request into SPEC.md. Use when the objective is clear but scope needs constraining.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Bound and de-risk a request into SPEC.md. Use when the objective is clear but scope needs constraining.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Optional engineering go/no-go on a plan. Use when execution safety needs review before implementation.
Implement approved plan slices. Use as the execute-stage entry point.
Sharpen a vague idea into a bounded objective. Use before framing when scope is undefined.
Build project truth from repo evidence. Use when steering is missing or stale.
Turn an approved spec into ordered slices. Use when framing is accepted and planning begins.
Recover active change and next action from artifacts. Use on fresh session with existing work.
| name | auto-frame |
| description | Bound and de-risk a request into SPEC.md. Use when the objective is clear but scope needs constraining. |
| metadata | {"stage":"frame"} |
Framing controller. Bounds and de-risks a request into a single SPEC.md.
First action: run node .agent/.automaton/scripts/get-context.mjs from the project root.
auto-frame produces the canonical artifact: SPEC.md when the request is frameable. SPEC.md is mandatory for frame completion; no file means no completed frame. It does not write code, create PLAN.md, or proceed to planning without a written spec. If the request needs discovery rather than one blocking decision, continue into auto-office-hours rather than writing a weak SPEC.
Loading discipline: hold the skeleton when present, the objective, constraints, risks, and source evidence needed to keep the spec real. Avoid exhaustive tree walks. When a lookup would otherwise pull wide reads into context, dispatch the read-only automaton-librarian (see .agent/.automaton/references/LIBRARIAN.md): it returns evidence, you keep the decision.
Artifact discipline: SPEC.md is the reloadable contract, not the whole dossier. Keep it compact enough to re-read; for large coherent work, summarize the contract and link normative detail under spec/*.md. One coherent outcome remains one spec even when it needs progressive disclosure.
Before finalizing SPEC.md:
references/quality.md when the spec feels broad, padded, or hard to verify.Read .agent/work/<active_change>/SPEC.md if office-hours seeded it. The skeleton is preferred context, not a prerequisite for framing. A legacy INTAKE.md from an earlier harness version is optional context when present. If neither exists, frame from the current request, conversation context, and repo evidence. Do not send the user back to office-hours solely because no skeleton exists.
Adopt settled office-hours context: work scale, work shape, Broader intent, target user or stakeholder, scope coverage, rejected framings, and anti-goals. Do not re-ask settled context or reintroduce rejected directions.
Read .agent/wiki/LEARNINGS.md when it exists: facts earlier changes paid to learn often bound feasibility, constraints, and risks before new discovery does.
State the goal in one sentence. If you cannot, ask one clarifying question. If the request still needs objective discovery or multiple material decisions before any useful SPEC can be written, continue into auto-office-hours's diagnostic flow in the same session.
If your SPEC would be narrower than the user's stated goal or office-hours broader intent, widen the SPEC, ask for confirmation, or record a Deferred / Not in scope note in this change's SPEC. The rule's home is .agent/.automaton/references/ARTIFACT-LIFECYCLE.md (Progressive Disclosure). Do not create ROADMAP.md phases from a narrowed SPEC; that decomposition belongs to auto-office-hours after the user approves it. Silent narrowing is a framing failure.
If the skeleton or conversation context includes scope coverage, compare the intended SPEC against each item before writing:
If no formal scope coverage exists but the request has multiple material asks, perspectives, constraints, or worries, build the lightweight check from available context. Do not drop a material item silently.
List only constraints, unknowns, and risks that change implementation or verification. Keep decision-critical material in SPEC.md; link larger coherent detail under spec/constraints.md, spec/risks.md, spec/gap-matrix.md, or similar. If constraints point to unrelated outcomes, ask which outcome to frame first.
Choose the minimum useful lenses from product, engineering, design, security, runtime. Default to product + engineering unless the request says otherwise. Read references/lens-selection.md when selection is not obvious.
If the change involves content creation - writing, article, brief, deck, blog post, newsletter, documentation, or similar prose - add the content lens and read references/content-framing.md.
Use this when the request needs discovery, not one blocking decision. Continue into auto-office-hours in the same session when the request lacks a problem, stakeholder, desired outcome, content audience/thesis, first independent outcome, or direction choice. Recommend auto-office-hours only when continuation is blocked by context pressure, host limits, or the user's choice to pause.
When this happens, follow auto-office-hours's contract and wait for the user to approve an approach before any artifact write. Do not complete SPEC.md until an approach is approved and frame-ready.
If a SPEC.md already exists for this change, read it and preserve all ## Review: sections. When office-hours seeded the skeleton, complete it in place: keep the approved sections faithful and add what the spec shape requires.
Do NOT proceed past this step without writing SPEC.md to .agent/work/<change>/SPEC.md.
Do NOT write SPEC.md while a needs-decision item would change scope, approach, or verification unless the user answers it or explicitly accepts an assumption.
Read references/spec-shape.md and write the SPEC with its core fields and conditional fields. Conditional fields appear only when their named trigger applies.
Apply the Artifact Signal Discipline rules from .agent/.automaton/references/FRAMEWORK.md while writing. For large coherent work, follow the spec layout in .agent/.automaton/references/ARTIFACT-LIFECYCLE.md (Progressive Disclosure).
If active_change is bootstrap or does not match the current objective, derive a new slug: YYYY-MM-DD-<kebab-case-objective> using today's date. Use that slug before writing SPEC.md.
After writing SPEC.md, run node .agent/.automaton/scripts/sync-status.mjs --active-change "<change>" --canonical-spec ".agent/work/<change>/SPEC.md" --stage frame from the project root. auto-plan owns the stage: plan mutation when planning begins, including inline continuation.
.agent/work/<change>/SPEC.md; canonical_spec and frame state recorded through sync-status.mjs.auto-office-hours's contract and do not report framing complete until an approved skeleton exists and SPEC.md can be completed.**Next:** auto-plan, <reason>. The user reading SPEC.md is the product review. If not frameable, continue inline into auto-office-hours with the concrete blocker.