菜单
Synthesize Architecture Decision Records from session transcripts. Extracts decisions via multi-agent pipeline and writes MADR files to a specified directory. Use after completing a /define or /do session to capture architectural decisions as durable records.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | adr |
| description | Synthesize Architecture Decision Records from session transcripts. Extracts decisions via multi-agent pipeline and writes MADR files to a specified directory. Use after completing a /define or /do session to capture architectural decisions as durable records. |
| argument-hint | <manifest-path> <output-dir> <transcript-path> |
| user-invocable | true |
Extract ADR-worthy decisions from a completed manifest workflow session and write individual MADR files. Operates as post-processing — runs AFTER the manifest workflow completes, not during it.
$ARGUMENTS = <manifest-path> <output-dir> <transcript-path>
All three are required positional arguments:
/define outputs this path at completion)If any required argument is missing: error and halt with usage:
Usage: /adr <manifest-path> <output-dir> <transcript-path>
Example: /adr ~/.manifest-dev/manifests/manifest-1234.md docs/adr/ ~/.claude/projects/.../session.jsonl
Spawn three extraction agents in parallel, each analyzing the session transcript through a different decision lens:
Architecture Lens — Identify technology choices, component structure decisions, integration approaches, and pattern selections. Look for: "we should use X", "the architecture is Y", structural decisions with alternatives discussed.
Trade-off Lens — Identify tensions where competing concerns were weighed. Look for: "A vs B", preference statements with reasoning, rejected approaches with "because", T-* items from the manifest's Approach section that trace back to transcript deliberation.
Scope & Constraints Lens — Identify deliberate inclusions/exclusions that shape system boundaries, key constraint decisions where alternatives existed. Look for: "out of scope", "we need to include", "the constraint is", INV-G* items from the manifest that arose from deliberation (not just mechanical quality gates).
Each extraction agent receives:
Each agent outputs a list of candidate decisions with:
A synthesis agent receives all candidates from Phase 1 and:
Deduplicates — Merge candidates that describe the same decision from different lenses. Prefer the version with richer context and alternatives.
Applies ADR-worthiness criteria — Read references/ADR_FORMAT.md in this skill's directory. Apply the decision test to each candidate: "Would a new team member joining in 6 months benefit from knowing WHY this was decided this way?" Remove candidates that fail.
Writes ADR files — For each surviving candidate, generate a MADR file at <output-dir>/YYYYMMDD-kebab-title.md using the template from references/ADR_FORMAT.md. Create the output directory if it doesn't exist. Use today's date for the YYYYMMDD prefix.
If the session transcript at <transcript-path> is unreadable or missing:
If no ADR-worthy decisions are found after synthesis:
On completion, output:
ADRs written to: <output-dir>/
| # | Title | File |
|---|-------|------|
| 1 | [title] | YYYYMMDD-kebab-title.md |
| 2 | [title] | YYYYMMDD-kebab-title.md |
Total: N ADR(s) from M candidate decisions.
/define runs, /adr processes the full transcript. ADRs may reflect decisions from any run in the session. Review output for relevance to your specific task.~/.claude/projects/<dir>/<id>.jsonl) is stable. If the format changes, extraction agents may need updating.End-to-end autonomous execution: figure-out → define → do, chained without manual approval gates. Use when you want to define and execute without intervention during planning, when the user asks for autonomous or end-to-end work, says just build it, or asks to tend or babysit a PR.
Figure things out together — any topic, problem, or idea. Presses relentlessly until shared understanding is reached. Use when you need to understand before acting, when figuring it out is the goal, or when the user asks to think through a decision, dig deeper, press an assumption, investigate why something is happening, or work through a problem.
Generate multi-CLI distribution packages from the Claude Code plugin. Converts shared skills and package assets for OpenCode, Codex CLI, and Pi under dist/. Run after changing plugin components to keep distributions in sync.
End-to-end autonomous execution: figure-out → define → do, chained without manual approval gates. Use when you want to define and execute without intervention during planning, when the user asks for autonomous or end-to-end work, says just build it, or asks to tend or babysit a PR.
Manifest executor. Works through Deliverables verifying every Acceptance Criterion and Global Invariant. Use when executing a manifest, running a plan, implementing a defined task, or when the user asks to run, execute, implement, or ship a manifest-backed plan.
Completion marker for the /do workflow. Outputs a plain-prose summary of what was built. Called by /do after every Acceptance Criterion and Global Invariant has fresh PASS evidence, when the manifest is complete, all criteria pass, or the workflow needs to wrap up with a completion summary.