| name | prompt-align |
| description | Compare an agent or skill prompt against its test harness skill for phase-model, skip-condition, and wiring consistency. Reports mismatches and identifies which file is authoritative. |
/prompt-align
Compare an agent/skill prompt against its test harness. Emit a mismatch report.
Invocation
/prompt-align <path-to-agent-or-skill> [--harness <path-to-harness-skill>]
If --harness is omitted, search for the harness automatically:
find "${CLAUDE_PLUGIN_ROOT}/.." -name "SKILL.md" | xargs grep -l "test.*<agent-basename>\|<agent-basename>.*test" 2>/dev/null
Step 1 — Read both files
Read the agent/skill file and the harness file in full.
Step 2 — Extract comparable structures
| Structure | Agent location | Harness location |
|---|
| Phase table (IDs + skip conditions) | L0 phase table | Step 4.3 skip table |
| Standard wiring | L1 wiring list | Standard wiring list |
| Phase names | Phase table Subject column | Phase Backlog row labels |
Step 3 — Compare and emit mismatch report
For each structure, diff the two versions:
### Phase table mismatches
| Phase ID | Agent says | Harness says | Verdict |
|----------|-----------|--------------|---------|
| P1 | Research (skip: self-contained) | Research (skip: ## What to do self-contained...) | ✓ equivalent |
| P6 | Tests + fix (skip: docs-only) | Tests + fix (skip: docs-only) | ✓ match |
| ...
### Wiring mismatches
(same table format)
### Phase name mismatches
(same table format)
Verdict values: ✓ match, ✓ equivalent (same semantics, different wording), ⚠ minor (semantically close but worth aligning), ✗ conflict (different behavior implied).
Step 4 — Authoritative source determination
State which file is authoritative for each conflict and why (usually the agent is authoritative; the harness is the consumer).
Step 5 — Emit fix list
For each ⚠ minor or ✗ conflict row, emit a one-line fix recommendation:
- Which file to change
- The exact string to replace or add
