Battam1111/Myco

Seven-round post-release audit of v0.5.9 surfaced four real bugs. This hotfix closes all four. Two audit rounds on the fixed tree confirm no new regressions. Zero contract-shape changes. Operators upgrade with pip install -U myco and nothing else.

Bugs fixed

1. SubstrateNotFound exit code silently downgraded — v0.5.8 promised exit 4 for "no substrate at this path" but build_context wrapped the exception in UsageError, downgrading to exit 3. Fixed: re-raise SubstrateNotFound with the helpful message. CI scripts that special-case exit == 4 now work.

2. Fresh-substrate lint noise — every freshly-germinated substrate reported 10 findings on first myco immune (5 MEDIUM SE1 + 1 MEDIUM SH1 + 4 LOW RL1) due to the canon template stamping kernel-path _ref fields. Fixed: template trimmed to minimal. Fresh substrate now reports 0 findings out of the box.

3. RL1 noisy on substrates without L1 protocol — RL1 fired 7 findings on every fresh substrate. Fixed: RL1 early-returns when docs/architecture/L1_CONTRACT/protocol.md is absent. The check is only meaningful for substrates that mirror Myco's L1 tree.

4. Canon JSON-Schema over-strict — v0.5.9 schema required subsystems.*.doc; Python validator didn't. IDE false positives. Fixed: schema matches kernel requirements.

Additional improvement

• .myco_state/unsafe_writes.log audit trail — every MYCO_ALLOW_UNSAFE_WRITE=1 bypass writes one line (<ISO-8601 UTC> <target>) to this log. Resolves a stale v0.5.9 TODO. Sets up the ledger a future SE-class dim will count from.

Seven-round audit summary

| Round | Scope | Result |
|---|---|---|
| 1 | Static gates (pytest / ruff / mypy / immune / hunger) | all green |
| 2 | E2E all 18 verbs on fresh substrate | all exit 0 |
| 3 | MCP server boot + tool registration | 18 verbs + 9 aliases = 27 tools |
| 4 | Static code (bare excepts, type-ignores, TODOs) | 1 stale TODO fixed (→ log) |
| 5 | Edge cases (empty canon, missing, corrupt) | exit codes 3/4/5 all correct |
| 6 | Fresh PyPI install | works; 25 dims discovered |
| 7 | Deep cross-check v0.5.9 canon on v0.5.10 kernel | backward-compat confirmed |

Gate status at release

• pytest: 757 passing (2 new tests: RL1 skip + unsafe_writes.log)
• ruff + ruff format: clean
• mypy src/myco: 0 errors
• myco immune on self: exit 0, 0 findings
• myco immune on fresh substrate: exit 0, 0 findings (was 10 at v0.5.9)
• myco hunger: clean
• python -m build + twine check: PASSED × 2
• jsonschema-validate live _canon.yaml: OK
• jsonschema-validate fresh canon: OK
• Exit-code matrix verified: 3 / 4 / 5 all preserved

Upgrading

pip install -U myco

That's it. Existing v0.5.9 substrates with the full canon shape continue to load and lint cleanly under v0.5.10. Fresh germinations get a minimal canon that produces 0 findings out of the box.