agentsclimarketplace
Sign in

Agent first planning

Skill avksp/agent-lifecycle-kit/skills/agent-first-planning

Create or revise production-ready SDD plans for agent execution, including clarification, specification, freeze inputs, ownership, budgets, and evidence contracts.From its SKILL.md

Install
npx -y skills add avksp/agent-lifecycle-kit --skill agent-first-planning

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 18 stars18 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.3 KB, as published. Nobody here has run it

Agent first planning

Use this skill when a request must become an agent-ready implementation plan before work starts. The output is a frozen-ready package, not implementation.

Contract

  1. Clarify only questions that can change scope, architecture, safety, acceptance, external authority, or write ownership. Record safe assumptions for everything else.
  2. Choose the SDD tier.
    • S0: one bounded mechanical change with one executable owner.
    • S1: one standard owner and no major architecture, security, performance, browser, data, or external-environment risk.
    • S2: multiple executable owners or meaningful architecture, security, performance, browser, data, release, migration, or external risk. Use the core tier resolve command when available, and never lower a tier without independent review.
  3. For S2, write a product specification first and require independent specification review before plan freeze.
  4. Produce a developer overview that explains the intent, user-visible result, architecture boundaries, risks, non-goals, and execution model.
  5. Produce a plan manifest or equivalent machine-readable contract with: requirements, acceptance criteria, evidence ids, workstreams, dependency DAG, exact write ownership, read-only inputs, lead-owned integration seams, validation commands, budgets, context limits, and final audit gates.
  6. Keep immutable planning authority separate from mutable run artifacts.
  7. Route the candidate plan to audit-agent-plan. Revise until it is READY_TO_FREEZE, explicitly blocked, or the review budget is exhausted.

When the CLI is available, run the structural gate before independent review:

agent-lifecycle plan completeness-check --manifest <plan.manifest.json>
agent-lifecycle plan check --manifest <plan.manifest.json> --require-completeness

The goal is compact-but-complete authority for the selected tier, not long prose. S0/S1 plans can stay short, but the fields required by their tier must be present.

Planning rules

  • Do not implement code while planning.
  • Do not assign shared integration to a worker by hiding it in another task.
  • Do not depend on chat history as authority when a plan package exists.
  • Do not claim production readiness without tests, evidence, rollback or release gates appropriate to the risk.
  • Keep plans small enough for workers: active packet, exact files, acceptance criteria, and evidence routes only.
  • For small-context hosts, require compact packet rendering with small-context-profile.v1; do not depend on full plan context.
  • Do not use shortness as a reason to omit required tier structure. Missing evidence routes, write ownership, budgets, context limits, or final gates are blockers when the selected tier requires them.

Output

Return these artifacts or their host-native equivalents:

  • clarification log and assumptions;
  • SDD product specification when required;
  • developer overview;
  • plan manifest;
  • plan lock or freeze candidate identity;
  • exact worker write-set manifest;
  • validation and evidence plan;
  • plan review request.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.