Fp Task

FP Task

Use this workflow for issue-backed implementation work in Switchyard.

Start

1. Inspect repo state with git status --short.
2. Load issue context:
   • fp context <id> or the exact context command the user gave.
   • fp issue get <id>.
   • If the issue names a parent, epic, dependency, spec, or ADR, load that too.
3. Establish a baseline before editing:
   • Run relevant scoped checks for the touched workspace when the issue scope is clear.
   • Run root checks for broad, cross-workspace, or unclear scope.
   • If a baseline check is too expensive or unrelated, record why it was skipped.
   • If the baseline is already failing, capture the failure before making changes.
4. Claim the task unless the user only asked for analysis:
   • fp issue update <id> --status in-progress.
   • fp comment <id> "Starting work: ...".

Understand

Map the issue into concrete deliverables before editing:

• Files to create or change.
• Commands and gates to run.
• Acceptance criteria and tracer bullets.
• Required fp comments, reflections, commit attachments, or status changes.
• Docs, drift, or spec backport obligations.
• Parent/child issue behavior, especially auto-close side effects.

Read linked specs and local docs before writing code. Let existing repo patterns decide structure.

Implement

• Keep changes scoped to the issue.
• Preserve unrelated dirty work.
• Use fp comment <id> "..." at meaningful milestones.
• If docs describe changed behavior, update docs in the same change and use the drift skill when bindings are involved.
• Prefer scoped checks while iterating. Inspect the workspace package.json first and run only scripts that exist. Use exact package names or path selectors, for example:
  • bun run --filter @switchyard/symphony-orchestrator test
  • bun run --filter ./playgrounds/symphony-daytona-playground typecheck

Review

Review is mandatory for implementation work unless the user explicitly opts out or the current environment cannot run subagents.

1. Ask a review subagent to inspect the diff for correctness, missed requirements, regressions, and test gaps.
2. If a review skill or plugin is available, instruct the subagent to use it.
3. Give the reviewer issue context, acceptance criteria, and the current diff. Do not ask for a rubber stamp.
4. Address actionable findings before final verification. If a finding is intentionally not addressed, explain why in the final fp comment.

If subagents are unavailable, state that exception clearly, perform a structured self-review, and ask for human or subagent-capable review before marking the issue done unless the user tells you to continue.

Verification

Before closing an implementation task, run root gates:

bun run test
bun run format:check
bun run check

Notes:

• bun run check runs oxlint, ast-grep, drift, and typecheck.
• bun run check does not run tests or format:check.
• Use scoped formatting during development. Run root bun run format only when root format:check must be made green, and call out formatter-only churn.

Completion Audit

Before marking done:

• Restate the issue as concrete requirements.
• Check each requirement against files, tests, command output, fp comments, and commits.
• Confirm fixtures or verifier suites actually cover the acceptance criteria.
• Treat missing evidence as incomplete.
• Confirm parent/epic status separately; do not let child completion imply the whole epic is done.

Commit And Close

1. Stage only the intended work.

2. Commit with the fp issue id in the message.

3. Post required reflection or final comment.

4. Mark done:

   fp issue update <id> --status done