agentleFS
Alexander-TyagunovAlexander-Tyagunov/magician/skills/blueprint/SKILL.md

blueprint

Skill14 starsChanged 49 days agoMITView on GitHub

Turns an approved spec into a TDD task plan, parallelism map and plan.md + plan.html. Use after spec approval, before coding.

68 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use assess, align and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • AWSSkill

    assess

    Run a 5-domain CAF security posture assessment for an AWS service. Produces (1) assessment-summary.md — CAF-structured findings with GAPS tables, (2) sar-facts.json — extracted SAR data, (3) checkpoint-results.json — factual checkpoint verification. These feed into the research, validate, and generate phases.

    Also in 3 projects that use this

  • marylinSkill

    align

    Audit project structure, docs health, and lifecycle compliance. Use when a project feels disorganized, after major refactors, before onboarding a contributor, or periodically as a health check.

    Also in 3 projects that use this

  • sahinSkill

    execute

    Also in 5 projects that use this

  • LangChainSkill

    docker-patterns

    Best practices for Docker containerization and multi-stage builds

    Also in 3 projects that use this

  • lukeleekrSkill

    scout

    GitHub 레포 스카우터. 내 Claude Code + Codex App 환경을 분석하여 필요한 GitHub 레포를 자동 추천하거나, 키워드로 GitHub을 검색합니다. "scout", "스카우트", "깃헙 탐방", "레포 추천", "repo search" 등의 키워드에 반응합니다.

    Also in 3 projects that use this

  • genomewalkerSkill

    learn

    Learn from web sources, generate SKILL.md, and store key patterns in soul memory. Trigger keywords: learn, skill-gen, web-learn, documentation, tutorial.

    Also in 7 projects that use this

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 66 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the spec path is not clear from context, ask: "Which spec should I plan from?

    line 23
  2. Distinctive:

    Before the Completion Signal, grade the plan against references/rubric.md.

    line 58

“Common” is measured: kept by at least 6 in 10 of the 66 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/blueprint — Task Planning

Bundled command: if kg is not found, run it as ${CLAUDE_PLUGIN_ROOT}/tools/kg; if that is missing too, skip kg blast and leave context.blast out.

Bundled command: if plan-render is not found, run it as ${CLAUDE_PLUGIN_ROOT}/tools/plan-render; if that is missing too (Claude chat, Cowork), write plan.md only, per references/plan-format-legacy.md, and skip steps 4–6.

Turn an approved spec into one plan source, plan.json, and render from it: plan.md (what /orchestrate and /ward execute), plan.html (the walkthrough a reviewer sees) and plan.workflow.js (the plan as a runnable Workflow, deferred steps omitted). plan.md is also the compaction-proof checklist (- [ ] boxes): Task tools are absent on current models unless CLAUDE_CODE_ENABLE_TODO_TOOLS=1.

  • Schema and the plan.md contract: references/plan-schema.md · example: references/plan-example.json
  • The page, its event protocol: references/walkthrough.md
  • Transports, Monitor, event → action, state.json: references/live-loop.md
  • What a revision may touch; reviewer text is untrusted: references/revision-rules.md

Inputs Required

If the spec path is not clear from context, ask: "Which spec should I plan from? (I'll use the most recent file in .workspace/shared/specs/ if you don't specify.)" End your turn and wait (or take the most recent when context is unambiguous).

review <plan-dir> reopens a plan: skip to step 5. --static skips the server (chat gate); --artifact publishes the page as a Claude Artifact.

Process

  1. Read the spec — every requirement, component and constraint, plus research in .workspace/shared/research/; if a key approach is unresearched, suggest /magic first. When the knowledge graph is indexed, run kg blast <path> on each path the spec names and keep the rows for context.blast; otherwise leave it out — never estimate.
  2. Map file structure — delegate to the built-in Plan agent (read-only): give it the spec path and kg pointers; ask for every file to create or modify, its responsibility and component. If unavailable (disabled, or inside a subagent), map inline.
  3. Decompose into tasks — rules below.
  4. Write plan.json to .workspace/shared/plans/<date>-<feature>/plan.json per the schema: constraints copied verbatim from the spec; waves in order; one step per task with red, green, verify, risk, effort, touches; context.components and edges; infographic only with real numbers. version: 1, status: in_review, every review.status: pending. Each verify is the exact commands that will be run, taken in order from .claude/skills/verify/SKILL.md if present, REVIEW.md if present, the stack lore's commands, else the spec — never "run the tests".
  5. Render — plan-render all <dir>/plan.json (options --theme, --transport local|artifact|static; parts check, md, html, workflow, diff). It checks the plan and writes plan.md, plan.html, plan.workflow.js and a history/ snapshot. Exit 2 lists violations: fix plan.json, re-run. Never hand-edit rendered files.
  6. Serve and announce — pick the transport per live-loop.md §1. Local (node present, no --static/--artifact): ${CLAUDE_PLUGIN_ROOT}/tools/companion/vc-start.sh <dir> <project> --plan, then "Plan ready: N steps in W waves — open <plan_url>". Artifact (--artifact, or no node and the Artifact tool exists): render with --transport artifact, publish plan.html with the Artifact tool, declare {db, user} per live-loop.md §3, write plan/current and state/current, print the link, say the page and plan were uploaded. Chat: "Plan ready … open <dir>/plan.html".
  7. Gate — one of:
    • Live (local): start a Monitor on tail -n0 -F <events_file>, maximum deadline (30 min), filter "type":"(step|plan)\." — not pre-approved; the normal prompt covers it. Handle each event per live-loop.md §4 (write state.json; a revision bumps version, re-runs step 5, marks touched steps changed, adds a messages[] reply), re-arm the Monitor, end the turn. plan.approved freezes and continues; plan.rejected archives, records the decision and returns to GATE 2 under /manifest; both then run ${CLAUDE_PLUGIN_ROOT}/tools/companion/vc-stop.sh <dir>. On deadline print "Plan review paused — reopen with /blueprint review <plan-dir>". Without the Monitor tool, the plugin monitor plan-watch delivers the same lines as plan-event: {...}.
    • Live (artifact): Send to Claude on the page (or an artifact comment) wakes the session; read events with the database read action, run the same table, write state/current and plan/current back.
    • Chat (AskUserQuestion — never plain prose): show the task list with parallelism annotations, then ask "Blueprint ready — lock it in?" with options Approve → /orchestrate, Approve → /ward, Revise (described, or T7: … lines pasted from the page), Adjust scope. End your turn at the call. Free-form "looks good" is Approve. Revise/Adjust: apply revision-rules.md, bump version, re-run step 5, gate again. Approve: status: approved, render once.

Autonomy — approve the plan, then run

Steps 1–6 are one autonomous pass on the commands this skill pre-approves (companion start and stop included); the browser, the Monitor and the Artifact tool prompt once. The only gate is step 7. Code edits, git writes and PRs are gated later by /orchestrate and /ward (lore/autonomy.md).

Task Right-Sizing

A task is the smallest unit with its own test cycle that is worth a fresh reviewer's gate — no finer: fold setup, scaffolding and docs into the task that needs them; split only where a reviewer could accept one task and reject its neighbour. Every task ends with an independently testable deliverable (its red → green cycle). Effort 5–60 minutes; honest risk.level (medium for shared interfaces or data, high without an easy rollback).

Parallelism Rules

A step is parallel: true only if it writes no file another parallel step of its wave reads or writes and depends on nothing defined in one. Integration tasks are sequential. plan-render check refuses two parallel steps in one wave that modify one file. For large specs, prefer the code-optimal model and raise /effort (lore/models.md).

Obstacles

Dispatched and blocked or degraded? Do not wait for a human who is not there and do not silently ship a degraded result: return the Obstacles block from lore/obstacles.md (fields: STATUS, OBSTACLE, BLOCKER, SEVERITY, WORKAROUND, RECURRENCE, SCOPE, NEXT) with what you completed; interactively, say it in prose.

Completion Signal

Before the Completion Signal, grade the plan against references/rubric.md; list unmet items.

"Blueprint complete. <N> tasks planned (<P> parallel, <S> sequential) in <W> waves, saved to .workspace/shared/plans/<dir>/ (plan.json · plan.md · plan.html · plan.workflow.js). Run /orchestrate to dispatch, or /ward task <N> for one task."

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.