fleet-methodology-guide

Stage awareness — what's expected at each methodology stage, what to do, what NOT to do, which tools and skills are available

SKILL.md

GitHub

Fleet Methodology Guide

Your Current Stage

Your stage is determined by task_readiness in your task context:

  • conversation (0-20%) — understand requirements with PO
  • analysis (20-50%) — examine what exists in the codebase
  • investigation (50-80%) — research options, explore approaches
  • reasoning (80-99%) — plan approach, reference verbatim requirement
  • work (99-100%) — execute the confirmed plan

Source of truth: config/methodology.yaml

Stage Quick Reference

conversation (readiness 0-20%)

Job: UNDERSTAND, not BUILD.

DO DON'T
Ask specific questions about unclear requirements Write code
Propose your understanding, accept correction Commit changes
Extract knowledge from PO Create PRs
Post draft proposals for PO review Produce finished deliverables

Tools available: fleet_read_context, fleet_chat, fleet_task_progress, fleet_artifact_create (drafts only) Tools BLOCKED: fleet_commit, fleet_task_complete, fleet_task_accept Skills to use: /brainstorming (mandatory before creative work), fleet-communicate

Advance when: PO confirms understanding. Readiness increases. Verbatim requirement populated. No open questions.


analysis (readiness 20-50%)

Job: UNDERSTAND WHAT EXISTS, not solve the problem.

DO DON'T
Read and examine the codebase Produce solutions (that's reasoning)
Reference SPECIFIC files and line numbers Write implementation code
Produce analysis_document artifact progressively Skip to planning
Present findings to PO via task comments

Tools available: fleet_read_context, fleet_commit (analysis docs only), fleet_artifact_create/update, fleet_task_progress Tools BLOCKED: fleet_task_complete Skills to use: fleet-gap, fleet-urls, /systematic-debugging, fleet-comment

Advance when: Analysis document covers relevant areas. PO reviewed findings. Implications for task are clear.


investigation (readiness 50-80%)

Job: EXPLORE OPTIONS, not decide.

DO DON'T
Research solutions, explore options Decide on an approach (that's reasoning)
Explore MULTIPLE options — not just the first one Write implementation code
Cite sources, compare tradeoffs Commit to a solution
Present findings to PO

Tools available: fleet_read_context, fleet_commit (investigation docs only), fleet_artifact_create/update, fleet_task_progress Tools BLOCKED: fleet_task_complete Skills to use: /brainstorming, /dispatching-parallel-agents, /adversarial-spec (architect), fleet-report

Advance when: Research document has multiple options. PO reviewed findings. Enough info to decide in reasoning.


reasoning (readiness 80-99%)

Job: PLAN, not execute.

DO DON'T
Decide approach based on requirements + analysis + investigation Start implementing
Produce plan that references verbatim requirement EXPLICITLY Skip contribution check
Specify target files and components
Map acceptance criteria to implementation steps
Present plan to PO for confirmation

Tools available: fleet_read_context, fleet_task_accept, fleet_task_create, fleet_commit (planning docs), fleet_artifact_create/update Tools BLOCKED: fleet_task_complete Skills to use: /writing-plans, fleet-plan, fleet-task-create, fleet-contribution (check inputs)

Before advancing: Verify required contributions received (use eng_contribution_check or pm_contribution_check). Missing contributions → fleet_request_input. Do NOT advance to work without required inputs.

Advance when: Plan exists and references verbatim. Plan specifies target files. PO confirmed. Readiness reaches 99-100%.


work (readiness 99-100%)

Job: EXECUTE THE PLAN, not redesign.

DO DON'T
Execute the confirmed plan exactly Deviate from the plan
Follow conventions: conventional commits, tests Add unrequested scope
Use pre-embedded task data (or fleet_read_context if injection:none) Modify files outside plan targets
Call fleet_task_accept with plan Skip tests
Call fleet_commit for each logical change
Call fleet_task_complete when done

Tools available: ALL tools Tools BLOCKED: none Skills to use: /test-driven-development, /verification-before-completion, fleet-commit, fleet-pr, fleet-contribution (consume inputs)

Required tool sequence:

  1. fleet_read_context — load task + methodology state
  2. fleet_task_accept(plan) — confirm plan (verbatim reference checked)
  3. fleet_commit(files, msg) — one or more, conventional format
  4. fleet_task_complete(summary) — push, PR, review chain fires

Readiness Is Not a Gate

Readiness defines your methodology stage. It does NOT block dispatch. The PO sets readiness to control what stage you operate in.

  • Readiness 10 → conversation stage → ask questions
  • Readiness 30 → analysis stage → examine codebase
  • Readiness 50 → investigation stage → research options
  • Readiness 80 → reasoning stage → plan approach
  • Readiness 99 → work stage → execute plan

Task Types and Required Stages

Type Required Stages
epic conversation → analysis → investigation → reasoning → work
story conversation → reasoning → work
task reasoning → work
subtask reasoning → work
bug analysis → reasoning → work
spike conversation → investigation → reasoning (no work)
blocker conversation → reasoning → work

Checkpoints

  • At 50% readiness → PO gets checkpoint notification (informational)
  • At 90% readiness → PO gate request (BLOCKING — only PO can approve)
  • At phase advancement → PO gate request (ALWAYS)