workflow

name	workflow
description	Use at session start or when unsure what to do next

Pipeline Profiles

Each feature has a pipeline profile. Check: ptsd feature show <id> --agent

Profile	Stages
full	PRD -> Seed -> BDD -> Tests -> Impl
standard	PRD -> BDD -> Tests -> Impl
lite	PRD -> Tests -> Impl

Context tells you what to do: ptsd context --agent Follow the action= field. Skip stages not in the feature's profile.

At each stage

Stage	Create skill	Review skill
PRD	write-prd	review-prd
Seed	write-seed	review-seed
BDD	write-bdd	review-bdd
Tests	write-tests	review-tests
Impl	write-impl	review-impl

Session protocol

1. Run ptsd context --agent -- see where each feature is and what to do next.
2. Pick the next feature/stage from the next: lines.
3. Apply the write- skill -> create artifacts.
4. Commit with [SCOPE] type: message format.
5. Run ptsd review <feature> <stage> <score> -- score 0-10, honest self-assessment.
6. Move to the next stage or feature.

Stage cycle (repeat for every feature × every stage)

write artifacts -> commit [SCOPE] -> ptsd review <feature> <stage> <score> -> next

Do NOT skip the ptsd review step. It records review verdicts. Without it the feature stays review: pending forever.

Gate rules

• No BDD without seed initialized (full pipeline only)
• No tests without BDD written (full/standard only)
• No impl without passing test review
• No stage advance without review score >= min_score (default 7)

Existing project (after ptsd adopt)

If you joined a project that was bootstrapped with ptsd adopt:

1. Run ptsd context --agent -- see which features exist and their stages.
2. Check for unmapped tests: ptsd status --agent -- if TESTS:0 but test files exist, they need mapping.
3. Check for missing PRDs: features without <!-- feature:<id> --> anchors in PRD.md will be blocked.
4. For features with passing tests but no BDD: set pipeline to lite with ptsd feature pipeline <id> lite.
5. For features that need better test coverage: keep standard pipeline, write BDD scenarios first.

Multiple blocked features

When ptsd context --agent shows several features blocked:

1. Start with features that have no dependencies on other features.
2. If features share state (e.g., shared storage format), design the data model in the first feature's seed/BDD, then reference it in dependent features.
3. Complete one feature fully before starting the next -- avoid spreading work across many features at once.

Common Mistakes

• Starting implementation without checking review-status.yaml first.
• Skipping the review skill after the create skill -- both are required at each stage.
• Forgetting to update review-status.yaml immediately after completing work.
• Working on a feature that is blocked by a gate (e.g., writing tests before BDD exists).