agentleFS
anivaranivar/agentic-sdlc/skills/plan-project/SKILL.md

plan-project

Skill0 starsChanged 37 days agoMITView on GitHub

Write one plan file for a project or a change: outcome with measures, decisions laid out as options for a decider, numbered assumptions each tied to the test that proves it, dated steps with an owner and a "done when", gates between phases, risks with triggers, a rollout with a rollback, a cut order and a never-cut list, and the commitments made along the way. The plan is written, shown, and the turn ends; implementation starts only after the decider approves. Triggers on: "write a plan", "project plan", "plan this project", "roadmap", "spec", "specify", "requirements", "before we build", "how would we approach", "PRD", "what are the steps".

7 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

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

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 1

    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:

    Every factual clause carries its tag and a number carries its source (EVID-1).

    line 99

“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

Plan a project

One plan file, in the template below, at docs/plan/PLAN.md unless the decider names another path. When the project has code, a repository AGENTS.md carries the rules that bind. Nothing else: no index, no brief, no mapping document unless asked.

The gate

Writing the plan and starting the work are different turns. Save the plan, summarise it, list the open questions, ask the decider to approve or amend, and end the turn. Planning tasks and writing code start only after approval arrives in a later turn.

The template

# <Project>: plan

Status: proposal | agreed · Delivery owner: <team or role> · Decides: <role> · Updated: <date>

Evidence tags: Verified (source opened), Measured (number plus query), Inferred, Planned.

## Outcome
What changes, for whom. 2 to 4 measures, each baseline → target, with how it is measured.
Not doing: <explicit non-goals>.

## Behaviour change
Breaks existing behaviour: yes / no. What changes exactly: <…>. Parked: <…>.
| Failure | Allowed impact | Mechanism that bounds it | Test that proves the bound |
| Client | Before | After | Compatible: yes / no / after <step> |

## Decisions
| # | Decision | Options | Proposal | Decides | By |

## Assumptions
| # | Assumption | Evidence | Proven by (test, query, vendor answer) | Status: open / proven / broken |

## Plan
### Phase <n>: <name>. Gate: <what must be true to move on>, reviewed by <role>.
Ladder: self-service <what>; then <role> within <time>; lanes that pass without the decider <which>; ceiling <role> within <time>
| ID | Step | Owner | Days | Depends on | Done when | Date | Status |

## Risks
| # | Risk | Trigger to watch | Mitigation | Owner |

## Rollout
Pilot → widen (1 → 5 → all, with a hold point at each). Rollback: <the one step that undoes it>.
Run: monitoring, runbook, escalation ladder (who, after how long).

## If time runs short
Cut in this order: <list>. Never cut: <list>.

## Commitments
| What | To whom | By | Status |

The rules

IdRuleIn practice
PLAN-1One plan file, dated steps with done-whenA step is at most three days, has one owner, a date the owner named, and a "done when" someone else can check. Each step is reversible, or says how it is undone. A gate sits between phases.
PLAN-2Assumptions tied to testsNumber each assumption, tie it to the test, query or vendor answer that proves it, or mark it open. The first step attacks the riskiest assumption. Vertical slices that work end to end, not layers.
PLAN-3Cut order and never-cut listSay what goes first if time runs short and what never goes. A slip is reported before its date; a scope cut goes in the headline; a baseline never moves silently.
PLAN-4Behaviour change stated up frontDoes it break existing behaviour, what exactly changes, what was parked. A failure-mode table (failure, allowed impact, mechanism, test) and a per-client compatibility table. New pieces degrade alone; one bug never stops the business.
PLAN-5Gates carry escalation laddersA self-service path, a named role, the lanes that pass without the decider, a ceiling, each with a response time.
PLAN-6No resourcing, trivia or blocker wordsNo headcount, no operational trivia, no alarm phrases (blocker, urgent, critical) in a repository document, no follow-up-ticket line where a decision belongs; severity lives in the tracker.
PLAN-7Decisions in the decider's wordsOptions in the table; the decision recorded in the decider's own words, with who decides and by when. The decider approves; each delivery step names a team or role as its owner.
PLAN-8Newest production versionCheck what the platform runs in production, then decide. Never design on a beta; never choose an older version to stay compatible with something not in production.
PLAN-9Platform changes prove out on stagingValidated there before production.

Sizing

A size with no basis is marked as a guess. Numbers carry their source: a query, file:line at a named commit, or a link (EVID-1).

Calibration, voice and the self-review

Every factual clause carries its tag and a number carries its source (EVID-1); the evidence words need evidence (VOICE-8); the plan states the current state only (VOICE-2); an open point stays open with who decides and by when (VOICE-9); the plan commits teams and roles, never the decider's own work (VOICE-13); the Decisions table is settled through grill-plan (GRILL-3). Before handing over, run the self-review in review-change (REVIEW-7) and one check of its own: the decider could explain the plan in three sentences without the document.

Repository AGENTS.md, when there is code

# <repo>: rules that bind
## How to work
- One plan step at a time. Update the plan with the code.
- Commit only when lint, typecheck and every test pass, gated on exit codes.
## Rules
1. <the invariants that must never break, each one line>

Companion skills

NeedSkill
Settling the decisions table before the plan is finalgrill-plan
Recording a decision that is hard to reversedecision-records
Turning approved steps into ticketscut-tickets
Landing each stepthin-slices

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.