AgentStack
SKILL verified MIT Self-run

Lean Plan

skill-nstungdev-claude-skills-lean-plan · by nstungdev

>

No reviews yet
0 installs
0 views
view→install

Install

$ agentstack add skill-nstungdev-claude-skills-lean-plan

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Lean Plan? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Lean Plan

Write a PLAN.md that a reviewer can approve or reject in 30–60 seconds, answering four questions: what gets built, how (briefly), what it affects, and what it explicitly won't do. It does not answer why or who — that's already in SPEC.md. Link to it, don't repeat it.

Written for a reviewing supervisor, not a line-by-line coder. The user delegates implementation to Claude and reviews checkpoints rather than reading code directly — so even though PLAN.md is more technical than SPEC.md, it must stay a fast approve/reject read. Full technical breakdown (step-by-step approach, data flow, rationale per decision) does NOT belong here — that level of detail is for lean-task once implementation actually starts. PLAN.md's "How" section is a 2–4 line summary of approach, not the full breakdown.

The Components table doubles as input for later task breakdown — each row must stand on its own well enough to become a task later, but the plan itself stays short.


Modes

Step 0 — Check current state

Check whether specs/{feature-name}/SPEC.md exists before doing anything else.

  • No SPEC.md → stop. A plan built on an unscoped feature just encodes assumptions as fact. Tell the user to run lean-spec first.
  • SPEC.md exists, no PLAN.mdinit
  • PLAN.md exists → update

Mode: init

Step 1 — Extract, don't interrogate

Read the feature's SPEC.md — it already carries the domain and user context; don't re-derive it here.

Extract what's already available from the conversation:

  • Components or modules already mentioned
  • The high-level approach, if already discussed
  • Existing systems/features/data likely to be touched
  • Anything already ruled out or marked as technically out of scope
  • Hard constraints already stated

Only ask about what is genuinely missing: the component breakdown, the approach, impact, and any hard constraints. Ask everything in a single message.

If a decision isn't final yet, don't write it down as one — send it back to SPEC.md's Open Questions instead.

Step 2 — Generate PLAN.md

Read references/plan-template.md and fill it in, using the same {feature-name} slug as the feature's SPEC.md.

Rules:

  • One page max, tables and short bullets only — the four review sections must read in under a minute
  • What (Components): one row per unit of work, with the action (create / modify / delete) and any key detail
  • How: 2–4 lines, plain language, high-level approach only — no options, no per-decision rationale, no step-by-step. If it's growing past 4 lines, that detail belongs in lean-task, not here
  • Impact: existing systems, files, features, or data this touches — this is what lets a reviewer judge blast radius without reading code
  • Won't Do: explicit technical exclusions, not soft preferences
  • Constraints: hard rules only

Step 3 — Confirm and save

Show the full generated PLAN.md to the user. Ask if anything needs adjusting. Once confirmed, save as specs/{feature-name}/PLAN.md — the same feature folder as its SPEC.md.


Mode: update

PLAN.md is a frozen blueprint, not a progress tracker — it records what was decided, not what's done. Once tasks are underway, only touch it if the technical design itself changes.

When to update

  • A component gets added, split, or removed
  • The approach changes
  • The blast radius (impact) turns out to be different than planned
  • A new hard constraint is discovered mid-build

How to update

  1. Read the existing PLAN.md
  2. Edit only the affected row or section — do not rewrite the whole file
  3. Append a row to the Revision Log with the absolute date (YYYY-MM-DD, resolved from the current date — never a relative term like "today") and a short description of what changed and why
  4. Trim the Revision Log to the 10 most recent entries
  5. Update Last updated at the top

If nothing about the technical design has actually changed, don't edit — progress alone (a component getting built) is not a reason to touch this file.


Principles

  • A blueprint, not a trackerPLAN.md records what was decided, not what's done; don't add status columns or checkboxes
  • Built on the spec, not a copy of it — link to SPEC.md, don't restate its problem, users, or scope
  • Written for the reviewer's 60 seconds, not the implementer's hour — the four review sections (What / How / Impact / Won't Do) are the whole point; anything that only serves implementation detail belongs in lean-task
  • "How" is a summary, not a spec — if it needs step-by-step or per-decision rationale to make sense, that's too much detail for this file
  • Decisions are final when written — an unresolved choice belongs in SPEC.md's Open Questions, not here
  • Components are future task units — write each row specific enough that it could be handed off as a task as-is, not vague enough to need re-interpreting later
  • Frozen once tasks start — edit only when the plan itself changes, never to reflect progress

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.