agentleFS
Sign inStart free
ltlongtmaltlongtma/solo-sdlc/skills/sdlc/SKILL.md

sdlc

Skill2 starsChanged 41 days agoMITView on GitHub

Use when a solo founder working with AI has a new product or feature idea, wants to brainstorm or validate a startup idea, starts a new project, says "run the SDLC" / "take this from idea to ship" / "validate my idea" / "critique this idea", or wants to resume an existing project by the process.

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    R — rubber-stamp approvals are removed: No plan approval, no PR sign-off.

    line 23
  2. Distinctive:

    Stop and ask the human only at a B gate.

    line 25
  3. Distinctive:

    No superpowers / gstack installed → do it by hand, with the same artifacts and gates.

    line 59
  4. Distinctive:

    No product code before the plan passes its gate (spikes in scratch only).

    line 63

“Common” is measured: kept by at least 6 in 10 of the 7 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

SDLC — one founder + AI

A 9-phase pipeline. The AI does the work and decides most things; the human answers only B items. Gates are scripts, not statements. This file is a router — read the reference for the phase you are in, not all of them.

When invoked

  1. Resuming a repo → follow the resume order in session-lifecycle.md: AGENTS.md → docs/status.md (next:) → python3 scripts/sdlc/check-plan.py <plan> --list (plan path from docs/status.md). Converge when it reports problems. The repo's docs/WORKFLOW.md wins over this skill where they disagree.
  2. Fresh idea, no repo → phases 0–1 in scratch; no git init until GO. See phase-0-1.md.
  3. Repo without the process → run scaffold.sh (idempotent), then enter at the real phase. See phase-0-1.md.
  4. Any repo → if docs/preferences.md has no confirmed ## Models table, ask the human once to pick models; keep working on the suggested defaults (model-tiers.md).

Decision policy

Full rules: decision-policy.md.

  • B — synchronous stop: money, pricing, spend over budget, GO/NO-GO, prod deploy, destructive data ops, legal or customer-facing content, cutting committed scope, brand. Write it to docs/gates.md with options, a default and blocks:; only blocked work waits.
  • T — the AI decides and records it in an ADR or the spec's Clarifications with reverse with: <word>.
  • R — rubber-stamp approvals are removed. No plan approval, no PR sign-off.

Stop and ask the human only at a B gate. (The one-time model question is asked without stopping.)

Phases

#PhaseReadGate
0Triage, choose trackphase-0-1.mdbacklog entry, track recorded
1Validate (new product)phase-0-1.mdproduct-critic report → B: GO/NO-GO
2 / 2.5Spec, Clarifyphase-2-spec.mdcheck-spec exit 0
3Architecturephase-3-arch.mdADR + docs/design/architecture.html
4Planphase-4-plan.mdcheck-plan exit 0 + tech-lead-reviewer, no unresolved BLOCKING
5Executephase-5-execute.mdper task: [A<n>] red → green, commit, plan ticked with hash
6QA / Reviewphase-6-verify.md → commands/gate.mdcheck-gate exit 0
7Releasephase-7-release.mdsmoke + rollback + runbook; B: prod deploy
8Retrophase-8-retro.mdlessons written

Tracks (trivial / feature / subsystem / new product) decide which phases run — see phase-0-1.md.

Gate = script

A gate passes when its script exits 0, never because a session says so:

  • scripts/sdlc/check-spec.py — spec IDs, Verify by, UI states.
  • scripts/sdlc/check-plan.py — task fields, coverage of every A-id, risk flags ⇒ Tier: strong, ticked hashes resolve.
  • scripts/sdlc/check-gate.py — passing [A<n>] tests, ledger, B answers, no unresolved BLOCKING. Required in CI.

Artifact formats the scripts parse: artifact-format.md.

How work runs

  • Superpowers prompts are pre-answered: brainstorming design/spec approvals are approved by policy unless the work contains a B item (phases 0 and 2).
  • Thin orchestrator, one fresh subagent per task, briefed with brief-template.md. Reports ≤15 lines.
  • State in the same commit as the code: plan tick with hash + docs/status.md (next:, waiting-on-human:). No "one phase = one session" — new session only at ~60% context at a task boundary, or after ship. See session-lifecycle.md.
  • Model tiers strong / standard / fast; risk flags force strong; the model changes only at spawn or escalation. See model-tiers.md. Optional skills and per-harness model mapping: integrations.md.
  • The author does not grade their own work — reviewers run in a fresh context. No subagent support → run each agent as a separate headless process.
  • No superpowers / gstack installed → do it by hand, with the same artifacts and gates.

Core rules

  • No product code before the plan passes its gate (spikes in scratch only).
  • Git is the source of truth; a ticked box without a commit hash is not done.
  • Diverging from the spec needs an ADR.
  • Technical claims are verified and cited in the plan's Verified research.
  • Retry ceilings: 3 for the same failure, 5 per task, then escalate (phase-5-execute.md).
  • A failed gate sends you back, not forward (session-lifecycle.md).

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.

Check your setup, free