Stats
Actions
Tags
From convergence
Execute outline phases with verification checkpoints. Use after /convergence-outline approval. Works through each phase sequentially: write tests first, implement, verify, commit. Stops and asks the human if 3+ fix attempts fail on any phase.
How this skill is triggered — by the user, by Claude, or both
Slash command
/convergence:implementThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Work through the outline phase by phase. Each phase follows TDD (test first, then implement) and must pass its verification checkpoint before proceeding.
Work through the outline phase by phase. Each phase follows TDD (test first, then implement) and must pass its verification checkpoint before proceeding.
1. NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST.
2. NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE.
3. IF 3+ FIXES FAIL: STOP AND ASK THE HUMAN.
/convergence-outline approval for structured features/convergence-design approval when skipping outline (smaller features)/convergence-debug identifies a fix (use Phase 4 directly)For each phase in the outline:
Write tests for this phase's expected behavior. Run them. They must fail because the feature doesn't exist yet.
If the test passes immediately, it's testing existing behavior. Fix the test.
Write the minimal code to make tests pass.
Clean up without changing behavior. Tests must stay green.
Run the verification command from the outline. Read the full output.
BEFORE claiming this phase is complete:
1. IDENTIFY: What command proves this phase works?
2. RUN: Execute the command (fresh, complete)
3. READ: Full output — check exit code, count failures
4. VERIFY: Does output confirm the claim?
- If NO: State actual status with evidence
- If YES: State claim WITH evidence
5. ONLY THEN: Claim the phase is complete
Red flags — if you catch yourself thinking any of these, STOP:
Commit with a descriptive message for this phase. One commit per phase.
If the checkpoint fails:
Run the full test suite. All tests must pass — not just the new ones.
| Claim | Requires | Not Sufficient |
|---|---|---|
| Tests pass | Test command output: 0 failures | Previous run, "should pass" |
| Phase complete | Checkpoint command: expected output | Code looks right |
| Build succeeds | Build command: exit 0 | Linter passing |
| Bug fixed | Symptom test: passes | Code changed, assumed fixed |
| All done | Full suite: 0 failures | Phase tests pass |
| Bad | Why | Do Instead |
|---|---|---|
| Writing code before tests | Can't prove tests catch real issues | RED first, always |
| Implementing all phases then testing | Horizontal — can't isolate failures | Verify after each phase |
| Skipping verification | "Confidence is not evidence" | Run the command, read the output |
| Attempt #4 without asking | 3 failures = wrong architecture | Stop, escalate to human |
| "While I'm here" refactoring | Scope creep, unrelated to outline | Stick to the outline |
| Keeping code written before tests | Biases your tests toward implementation | Delete it, start with tests |
npx claudepluginhub c-sonnier/convergence --plugin convergenceGuides task implementation via Conductor's TDD workflow: mark plan.md progress, red-green-refactor cycle with pytest coverage checks, git commits, and verification protocols.
Executes TDD DEV stage: verifies failing tests exist, reads tests and design, implements minimal code one test at a time until all pass, verifies build, commits.
Executes implementation plans phase-by-phase: dispatches subagents per task, reviews once per phase with code-review skill, loads phases just-in-time, prints full outputs for transparency.