SECURITY REVIEW
Not yet assessed
Review the original instructions and requested permissions before installing.
No security review is available for this catalog entry yet.
Testing · Automation / AGENT SKILL
heal
0
Repairs failing Mumei source through a self-healing loop driven by verifier feedback and LLM attempts.
Run the mumei-agent self-healing loop on failing .mm source using mumei verify JSON feedback and LLM repair attempts.
Understand the trade-offs.
SKILL QUALITY
Not yet assessed
How clearly the skill guides your agent, how complete its workflow is, and how you can check the outcome.
No quality assessment is available for this catalog entry yet.
The full skill.
Original instructions from the publisher’s SKILL.md
Given a `.mm` source file and optional verifier error report, repair it through the self-healing loop until Mumei verification succeeds or retry limits are exhausted.
# Step 1: Prepare source and error report
Action:
Locate the `.mm` source file. If an error report is provided, keep it alongside the source; otherwise the agent will run `mumei verify --json`.
Expectation:
The file exists, is safe to modify, and the environment has `LLM_API_KEY` or `OPENAI_API_KEY` for non-dry-run repair.
Result:
If inputs and LLM config are ready, proceed to Step 2.
# Step 2: Run the self-healing loop
Action:
Invoke `uv run python -m agent heal`, optionally selecting retry count and strategy.
Expectation:
The loop backs up the original file, runs Mumei verification, asks the LLM for a repair using structured feedback, writes the candidate, and re-verifies.
Result:
The command exits successfully when the repaired file verifies.
```bash
uv run python -m agent heal input.mm --max-retries 3
uv run python -m agent heal input.mm --strategy multi-stage
```
MCP equivalent:
```text
heal_file(source_code, error_report)
```
# Step 3: Confirm repair
Action:
Re-run Mumei verification directly and inspect the diff.
Expectation:
`mumei verify --json` reports success. The patch is minimal and preserves intended contracts.
Result:
Report the fix, backup path, and verification evidence.
```bash
mumei verify input.mm --json
```
# Parameters
| Parameter | Type | Required | Default | Description |
|-----------|------|----------|---------|-------------|
| source_file | path | no | `examples/sword_test.mm` | `.mm` file to repair |
| error_report | JSON/string | no | | Existing verification report |
| max_retries | int | no | config default | Maximum repair attempts |
| strategy | string | no | `AGENT_STRATEGY`/`single` | `single` or `multi-stage` |