commands

Instruction file imported from Code-Munkiz/ham (.cursor/rules/commands.mdc). Copyright stays with the author.

Ham Commands

/audit-context-engine

Run the Context Engine Auditor checklist against src/memory_heist.py.

1. Read src/memory_heist.py.
2. Walk the Context Engine Hardening skill checklist (.cursor/skills/context-engine-hardening/SKILL.md), including continuation/parser marker coupling.
3. Report findings as a table: item, status (pass/fail), detail.
4. Do not auto-fix. Present findings for review.

/wire-agent-context

Wire shared repo context into Hermes-supervised role prompts in src/swarm_agency.py (context assembly only; no CrewAI).

1. Read the Agent Context Wiring skill (.cursor/skills/agent-context-wiring/SKILL.md).
2. Read src/swarm_agency.py and src/memory_heist.py.
3. Add from src.memory_heist import ContextBuilder (or equivalent) if missing.
4. Single discovery pass: build one ProjectContext (or one ContextBuilder that shares a single discovered context) and vary only render budgets per agent. Do not create multiple ContextBuilder() instances that each run a full ProjectContext.discover() (duplicate scans and git calls).
5. Prefer loading per-agent budgets from .ham.json / config (discover_config) over long-term hardcoded magic numbers; use code defaults only as fallback.
6. Inject per-role rendered context strings into each active role backstory/prompt surface.
7. Run lints. Present diff for review. Update VISION.md status if wiring completes a milestone.

/review-role-boundaries

Audit all agent definitions for role boundary violations.

1. Read the Role Boundaries rule (.cursor/rules/role-boundaries.mdc).
2. Read src/swarm_agency.py.
3. For each agent, check: does its goal, tools, and task assignments stay within its charter?
4. Report violations as a table: agent, field, violation, suggested fix.

/test-context-regressions

Generate or run regression tests for src/memory_heist.py.

1. Read the Repo Context Regression Testing skill (.cursor/skills/repo-context-regression-testing/SKILL.md).
2. If tests/test_memory_heist.py exists, read it and check coverage against the skill's test categories.
3. If it does not exist, generate it with pytest + tmp_path fixtures covering all 6 categories.
4. Run pytest tests/test_memory_heist.py -v and report results.

/rebrand-memory-heist

Scan src/memory_heist.py for any remaining Claude/Claw references and replace with Ham equivalents.

1. Read src/memory_heist.py.
2. Grep for claude, Claude, CLAUDE, claw, Claw, stolen, Stolen (case-insensitive).
3. If matches found, apply replacements per the memory-heist-conventions rule.
4. If no matches, report clean.

/refresh-swarm-contract

Re-validate the entire swarm against VISION.md.

1. Read VISION.md.
2. Read all pillar modules: src/swarm_agency.py, src/tools/droid_executor.py, src/hermes_feedback.py, src/memory_heist.py, src/llm_client.py.
3. For each pillar, check: does the module's current code match the role described in VISION.md?
4. Check: is memory_heist.py imported and used by the active orchestration path (currently swarm_agency.py)?
5. Check: do current role definitions respect supervisory-vs-execution boundaries?
6. Report as a table: pillar, module, vision role, actual status, gaps.
7. If the table fixes factual drift, apply updates to VISION.md (status table and next milestone) per the vision-sync rule.