AGENT SKILL

pmo-intake

Use when turning source material into scoped, source-referenced PMO requirements and intake decisions.

License 许可
MIT
最近更新
2026-10-14
Stars
★ 3
Forks
1

01 前言 Front Matter

name
pmo-intake
description
Use when turning source material into scoped, source-referenced PMO requirements and intake decisions.

03 SKILL.md 正文

pmo-intake

Purpose

Convert source material into confirmed requirements, assumptions, questions, scope, and mode recommendations.

Trigger

Use for new source review, requirement extraction, gap analysis, scope clarification, or source conflict triage.

Required Inputs

PROJECT.md, relevant source/**, and any user-stated goal. Read only the context set allowed by CONTEXT-ROUTER.md and pmo-config/context-map.json.

Allowed Context

Default to PROJECT.md plus the specific source files needed. Do not bulk-load archived skills or every source file unless the task is an impact analysis or conflict review.

Onboarding

Before any intake work on a project whose PROJECT.md does not yet declare both Execution path: and Default mode:, ask the user directly. Do not infer, detect, or silently default either answer — docs/concepts/execution-paths.md and DEC-011 both treat a system claiming to have "detected" either answer, rather than stating what the user declared, as a fabricated evidence claim. This mirrors axiom init's interactive wizard (cli/axiom.mjs) for the skill-based flow, which never invokes that CLI.

Ask, in the user's own words if useful, but cover both:

  1. Who builds this work — a developer or vendor after a verified handoff (development_handoff), or an AI execution agent under a governed contract, verified afterward against git ground truth (governed_ai_execution)? See docs/concepts/execution-paths.md.
  2. How strictly should this be governed — Lite, Standard, or Strict? If the user is unsure, walk through pmo-config/policy.json enums.strict_triggers (payment, financial calculation, PII, sensitive data, authentication, authorization, permission, irreversible action, external integration, compliance, production migration, critical infrastructure, public-sector formal acceptance) — any match recommends Strict.

Record both answers in PROJECT.md (Execution path: and Default mode: lines) and log the choice in decision-log.md if the project keeps one, stating plainly that this is what the user declared. Skip this step only when PROJECT.md already carries both declarations from a prior session — re-asking on every message is not required. A project may still escalate a specific work item to Strict later via the normal per-item mode recommendation below; this onboarding step sets the project-level default, not a ceiling.

Mode Behavior

Use the mode, strict-trigger, evidence-status, and approval enums from pmo-config/policy.json; do not repeat or extend them inside the skill. Lite captures only essential source, scope, and release approval needs. Standard captures normal requirements and design implications. Strict escalation follows the configured strict triggers.

Execution Steps

  1. Complete Onboarding above if PROJECT.md does not yet declare both Execution path: and Default mode:.
  2. Identify confirmed facts, assumptions, and open questions.
  3. Assign stable requirement IDs and source_ref.
  4. Mark evidence status from configured policy values.
  5. Recommend Lite, Standard, or Strict per work item.
  6. Update or request Source Snapshot refresh when source files change.
  7. When the project declares Research mode guided or auto, draft the Research Brief from preliminary requirements and unknowns (see docs/concepts/research-workflow.md); keep every material claim source-backed in RESEARCH/PROVENANCE.json. Research findings are candidate evidence that only a Human decision at Scope can act on.

Output Contract

Return confirmed requirements, assumptions, open questions, source references, confidence notes, and mode recommendation.

Approval Rules

Scope approval is human-owned. AI may draft rows but must not mark approval as granted without explicit source evidence.

Validation Command

node cli/axiom.mjs validate --project <project> --mode <mode> --gate Scope

Prohibited Actions

Do not invent requirements, bulk-load all source by default, silently resolve conflicts, push commits, deploy, or approve release. Do not write or infer Execution path: or Default mode: into PROJECT.md without first asking the user per Onboarding above, and never describe either as "detected."

Completion Criteria

Every in-scope requirement has source_ref, evidence status, and either confirmed acceptance or an open question.

04 同仓库更多 Skills