Battam1111/Myco

v0.8.6 - 2026-05-12 - Dead-dim revival sweep + 永恒删减 retirement wave

10-way parallel-agent audit (Round 7 of the v0.8.x cleanup sequence)
uncovered eight lint dimensions that had been emitting zero findings
for entire release windows because of hardcoded paths that no longer
matched the live substrate layout. This molt simultaneously revives
those gates AND retires three permanently-silent dimensions whose
infrastructure premise was never built.

What changed

Dim count: 50 → 47 (net −3 from the SE4/RL2/RL3 retirement). The
new totals — Mechanical 32 + Shipped 2 + Metabolic 6 + Semantic 7 —
match the live dimensions_run output of myco immune.

Real bugs fixed (8 dims newly emitting again):

• CL2 (oauth token residency): path src/myco/surface/mcp_auth.py
  → src/myco/boundary/surface/mcp_auth.py. CL2 had silently
  returned for every release v0.6.0…v0.8.5 — token-redaction
  enforcement was a permanent no-op for five minor versions.
• MF5 (generated mirror integrity): bundle dirs agents/,
  commands/ → .plugin/agents/, .plugin/commands/. The bundle
  paths moved at v0.8.4 (root cleanup); MF5 silently no-op'd until
  this fix.
• DI2 (discipline hooks content): hooks/hooks.json →
  .plugin/hooks/hooks.json (the actual Cowork-bundle binding).
• M1 (canon identity fields) fix(): root / "_canon.yaml" →
  ctx.substrate.paths.canon (canon-configurable layout support).
• CG1 (doctrine ← src reference): hardcoded
  "docs/architecture/L2_DOCTRINE" → ctx.substrate.paths.docs
  with the subpath joined locally.
• CG2 (subpackage → doctrine link): regex extended to accept
  .docs/ + .myco/notes/ prefixes (hidden-prefix layout support).
• SE5 (version anchor freshness): literal MYCO.md + _canon.yaml
  globs replaced by ctx.substrate.canon.entry_point and
  ctx.substrate.paths.canon lookups.
• digestion/pipeline.py ordering: check_write_allowed now
  runs BEFORE integrated_dir.mkdir(...) (same class of bug as
  the v0.8.5 molt-then-write reorder).

Coverage gap fixed:

• LB2 (Living Bets two-regime axis) was registered in _BUILT_IN
  but missing from pyproject.toml::[project.entry-points."myco.dimensions"]
  and .myco/canon_lint.yaml::dimensions. The dim was loading only
  through the gap-fill fallback; now declared explicitly in both
  inventories. metrics.lint_dim_count corrected 51 → 47.

Cross-platform consistency:

• ingestion/adapters/code_repo.py::IngestResult.source now POSIX-
  normalizes via .as_posix() so Windows ingestion sources don't
  embed backslashes that break cross-platform search/dedup.

Excretions (per L0 P3 永恒删减):

• SE4 (reciprocal backlink): white-list shipped permanently
  empty at v0.6.0; never populated through five releases.
• RL2 + RL3 (R3 sense + R4 eat discipline signals): read
  .myco/state/session_calls.jsonl that no production code has
  ever written — dead-letter checkers from landing.
• session_end_run re-export in myco/cycle/__init__.py:
  v0.5.x → v0.6.0 rename horizon shim; satisfies the doctrine note
  that scheduled deletion "via a v0.9+ craft once the rename horizon
  is doctrinally past".
• _SKIP_DIRS clone + _iter_py_files in circulation/graph_src.py:
  divergent skip-dir set + duplicate walker. Now delegates to the
  canonical core/skip_dirs.should_skip_dir.
• mcp-resources pyproject extra: orphan since the v0.8.5
  boundary/surface/mcp_resources excretion. Removed.
• .scripts/__pycache__/install_cowork_plugin.cpython-313.pyc:
  orphan bytecode. Removed.

Doctrine drift fixed:

• .docs/INSTALL.md: 6 .plugin bundle extension references →
  .zip (Anthropic GitHub issue #40414 / v0.7.4 hotfix).
• .docs/architecture/L2_DOCTRINE/extensibility.md: "Per-host"
  section rewritten to reference the data-driven
  boundary/install/clients.py::JsonClientSpec table; the
  excreted boundary/host_integration/ package + retired MF3
  dim acknowledged as v0.8.5 永恒删减 entries.
• .docs/architecture/L3_IMPLEMENTATION/package_map.md: dim totals
  • subcategory membership refreshed to match the v0.8.6 roster;
    excretion log gains host_integration v0.8.5 retirement +
    risk_classifier.py v0.8.5 retirement.
• .docs/architecture/README.md: title v0.8.5 → v0.8.6; dim count
  51 → 47; subcategory breakdown corrected.
• .docs/README.md: extensibility-axis ref points at
  boundary/install/clients.py (not the excreted symbionts/);
  verb-surface SSoT path bumped to boundary/surface/manifest.yaml.

Example substrate canons (v0.5.10 schema-v1 → v0.8.6 schema-v3):

• .docs/examples/minimal/_canon.yaml and
  .docs/examples/research-assistant/_canon.yaml rewritten to the
  current shape: llm_policy enum (replacing no_llm_in_substrate),
  cycle + boundary subsystem rows, governance.last_living_bets_audit_at,
  metrics.lint_dim_count: 47. A downstream user copying either as
  a template now passes SC1 schema-parity on first myco immune.

Config cleanup:

• .claude/settings.local.json: stripped deprecated verb aliases
  (session-end, reflect, distill, perfuse, genesis) and
  cross-project permission entries. Rewritten to list only the
  current 20-verb surface.
• .gitignore: tests/benchmark/.cache/ → .tests/benchmark/.cache/;
  added .claude/worktrees/ to keep abrupt-termination cruft from
  being committed.
• .docker/.dockerignore: .cowork-plugin exclude removed (directory
  consolidated into .plugin/ at v0.8.5).
• pyproject.toml: sdist include gains /.myco (canon-tree
  skeleton ships in sdist now); matching exclude gains
  /.myco/state/** (runtime state doesn't pollute the distribution).
• .scripts/bump_version.py: scripts/sync_plugin_mirrors.py →
  .scripts/sync_plugin_mirrors.py (path missed by the v0.8.4
  hidden-prefix root cleanup; would have failed for any
  Myco-self bump until this fix landed).

Break from v0.8.5

No backward-incompatible contract change. Three dims (SE4, RL2, RL3)
disappear from myco immune --list output — substrates that pinned
those IDs in their lint.skeleton_downgrade.affected_dimensions
allowlist will see an unknown-ID warning at load (existing kernel
behavior; not a hard failure). Eight dims (CL2, MF5, DI2, M1 fix,
CG1, CG2, SE5, plus LB2 via entry-points registration) now produce
findings on substrates where they previously emitted none — this is
the intended behavior; new findings on substrates that were silently
missing checks are signal, not regression.