ADR-007 + ADR-008 — runtime lifecycle foundation - #125
Draft
AreteDriver wants to merge 40 commits into
Draft
Conversation
…tern) ADR-007 establishes a single systemd target (animus-runtime.target) as the Animus lifecycle boundary. Three deployment profiles — development-local (default), desktop-login, continuous-node — bind the target to different host targets via add-wants symlinks. KillMode=control-group, Delegate=no, and PartOf= on every service unit are mandatory. A four-state ProcessClassification with provenance rules (registry identity + ≥2 independent proofs + UID match) replaces the existing recovery model; pgrep is explicitly disallowed for any authoritative classification. ADR-008 is the seven-step adversarial review pattern that produced these decisions. Status moves from Proposed to Accepted; the implementation is separately tracked in docs/specifications/animus-runtime-lifecycle-build-spec.md and the lifecycle package in packages/bootstrap. Refs ADL-20260804-001
20-section spec covering purpose, current state, target architecture, ownership matrix, deployment profile matrix, unit-file design, the 16-step atomic profile-switch transaction, health contract, process provenance model, control app, tray behavior, dashboard API changes, installer/migration, rollback, security boundaries, test architecture (20 required tests), release stages, acceptance criteria, post-implementation audit, and cross-references. The spec is the implementation contract for the lifecycle foundation package and the test harness. The 20-test matrix is fully addressed by the 54-test suite in tests/test_runtime_lifecycle/. Refs ADR-007, ADR-008
New pure-function package animus_bootstrap.lifecycle implementing: - classification.py — ProcessClassification (Managed / Recoverable / Orphaned / Unknown) with provenance rules. Orphaned requires registry identity + cgroup_alive OR ≥2 independent proofs + UID match. UID mismatch disqualifies. Recovery path requires one of executable, cmdline, start-time. No pgrep anywhere. - health.py — HealthState (7-state: OFFLINE / STARTING / HEALTHY / DEGRADED / FAILED / STOPPING / UNKNOWN). HealthContract is versioned (schema_version: '1'), strictly validated, and parse() rejects unknown states, negative counts, and missing timezone. - profile.py — ProfileSwitcher with a 16-step atomic switch transaction (drop-ins → daemon-reload → add/remove wants → verify → persist profile.json). Verification checks MemoryMax, KillMode, CPUQuota, and Delegate on the daemon's effective state. continuous-node requires user_consent=True. Any failure rolls back; the rollback's daemon-reload is logged but not fatal. - systemd.py — SystemdStateReader wrapping systemctl --user show into typed UnitState dataclasses. parse_show_output handles KEY=VALUE output strictly. All four modules are pure; the test harness exercises them via the FakeSystemd backend without touching the live user manager. Refs ADR-007, ADR-008, build-spec §3-§11
54 tests in tests/test_runtime_lifecycle/ covering all 20 cases from the build spec §16 matrix: - test_animus_runtime_target.py — target lifecycle, PartOf vs Wants, tray isolation, drop-in effective properties - test_stray_classification.py — 4-state classification boundary tests including UID mismatch, cgroup_alive decisive, and decisive-proof thresholds - test_health_state.py — 7-state derivation, contract round-trip and strict validation (unknown state, negative counts, missing timezone) - test_profile_switching.py — atomic switch, obsolete symlink removal, rollback, continuous-node user_consent guard, development-local no-binding - test_no_pgrep_in_lifecycle.py — AST-based static check that pgrep, pkill, kill, and signal are never called from authoritative classification paths - test_harness_cleanup.py — XDG fixtures, unique test prefixes, free-port allocation, no live-systemd reach - test_exclusions.py — KillMode=process and Delegate=yes are absent from the lifecycle source The harness uses a FakeSystemd that records method calls and synthesizes Wants= / drop-in effective properties without invoking systemctl --user. XDG_CONFIG_HOME and XDG_RUNTIME_DIR are monkey-patched to tmp_path; the live runtime cannot be touched. The package was renamed from test_runtime/ to test_runtime_lifecycle/ because the existing tests/test_runtime.py covers the AnimusRuntime orchestrator and the two would collide during pytest collection. Refs ADR-007, ADR-008, build-spec §16
Three new operator-facing documents: - docs/systemd/animus-runtime.md — canonical target unit, canonical service unit (KillMode=control-group, Delegate=no, PartOf=), profile switching, health contract, lingering policy. The 9-step manual switch procedure is documented for advanced operators who command systemctl by hand; the control app is the supported path. - docs/operations/process-registry.md — the four-state ProcessClassification, the six proof kinds, the registry, the cleanup CLI, the no-pgrep rule and why it exists, and the posture toward Unknown (do nothing, expose data, let the operator decide). - docs/reviews/animus-runtime-lifecycle-four-lens-review.md — the Phase 8 adversarial review across architect, Linux/systemd specialist, reliability engineer, and security/red-team lenses. 19 findings; 15 closed in-review, 4 tracked as Phase 9 followups (rollback daemon-reload warning, expanded verification properties, consent log path, drop-in directory permissions). Refs ADR-007, ADR-008
Root CLAUDE.md: a one-line addition to the Bootstrap layer overview naming the Phase 6 lifecycle foundation and pointing to the build spec + operator guides. packages/bootstrap/CLAUDE.md: a new 'lifecycle/' entry in the package tree, and a 'Runtime Lifecycle (Phase 6)' section in the anti-patterns block explicitly forbidding: - pgrep / pkill / kill from authoritative classification paths - Orphaned claims from /proc alone (registry identity first) - silent user-lingering enable - live-runtime mutation from tests (FakeSystemd only) - non-atomic profile switches - any service unit missing KillMode=control-group + Delegate=no + PartOf=animus-runtime.target Refs ADR-007, ADR-008
- 6 atomic commits on docs/adr-007-008 (none on main) - 54-test lifecycle harness; 139/139 passing in focused suite - 20/20 spec test matrix coverage - 19 four-lens findings; 17 closed, 2 Phase 7 followups - Hard-constraint audit: every hard constraint honored - Pre-existing 42 failures in full bootstrap suite confirmed unrelated (test-order interaction in dashboard tests)
- Replace 'pre-existing and unrelated' claim with 'existing full-suite order-interaction failures not reproduced in the focused lifecycle suite' - Add the attribution caveat: an origin/main baseline run is the appropriate followup to make the attribution defensible - Add an attribution followup row to the open-items table The lifecycle work does not run an attribution comparison against origin/main; the claim is therefore not proven, only strongly suggested by the focused-suite isolation behavior.
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
CodeQL found more than 20 potential problems in the proposed changes. Check the Files changed tab for more details.
Adds animus_forge.governor adapter package (ADL-20260805-001): subprocess-only bridge to the alg CLI; never imports animus_loop_governor.* (vendor-independence preserved). Public seam: - GovernorAdapter.ensure_run(repository, mission_id, contract_path, known_run_id=None) — strict 3-step resolution: validate known run id → filesystem hint under .animus-loop-governor/runs/ → alg start. Cross-mission mismatch on Step 2 silently falls through to Step 3. - GovernorVerifierCitizen runs alg verify after the worker chain, mapping rc 0/3/infrastructure failures to CitizenOutput.status (completed / needs_repair / failed). - MissionScheduler._start_ready_mission() gates READY→RUNNING on ensure_run() success; failure leaves mission READY for the next tick (no BLOCKED status in the enum). Persists mission.metadata['governor_run'] atomically with the transition. Exit-code contract: rc 4-98 → GovernorError, rc >=99 → RuntimeError (fail loud for impossible codes). Sanitized env strips ANTHROPIC_API_KEY, OPENAI_API_KEY; explicit shell=False in subprocess.run. Local constraints (forge/CLAUDE.md): p95=77, pathlib.Path, no bare except, no print() logging, no Quorum dep, stdlib + existing Forge only. Tests: 112 passing, adapter.py 100% covered, total 98.63% (>=97% gate). All subprocess calls bypassed via FakeGovernorClient test double; autouse _isolate_alg_path strips PATH unless ANIMUS_LOOP_GOVERNOR_INTEGRATION=1. Co-Authored-By: Claude <noreply@anthropic.com>
Collaborator
Author
Heads up — PR body needs manual restorationThe PR body was accidentally overwritten with a literal file path What landed on this branch (commit
|
…ion smoke The parser previously anchored on the path line's leaf (`lines[1].name`), which broke when Rich soft-wrapped the run id across multiple lines — reproduced with `COLUMNS=5` on the animus-loop-governor 0.1.0 wheel. The fix anchors on the canonical `Created run <id>` marker; whitespace collapse + ANSI/Rich-markup stripping handle every wrapping pattern. Pinned by 4 new unit tests (wrapped long path, ANSI escapes, Rich markup tags, missing marker) and a 5-test real-binary integration suite at `tests/test_governor/test_integration.py` gated by `ANIMUS_LOOP_GOVERNOR_INTEGRATION=1`. All 5 integration tests pass locally against the installed wheel. Closes the third item of ADL-20260805-001: real-binary integration proven, not just `FakeGovernorClient`. Coverage: 98% (target ≥97%).
New `Run loop-governor integration tests` step in the test-forge job, gated on `push` to `main` so PR branches don't block on PyPI wheel availability. Pinned to `animus-loop-governor==0.1.0` (ADL-20260805-001). When the wheel is unavailable the suite skips gracefully — no CI failure, preserving offline-development ergonomics. Completes ADL-20260805-001 follow-up #3 (CI wheel integration).
test_mission_delete_cascades_to_tasks set PRAGMA foreign_keys=ON inside a transaction body. Per SQLite docs, the pragma is a no-op within a transaction — it must be set on the connection before any BEGIN. The test silently bypassed FK enforcement and the orphan task was never deleted, causing the assertion to fail. Fix: issue the PRAGMA outside the transaction wrapper. The pragma is connection-scoped, so it persists for the lifetime of the connection. Re-enable then disable around the DELETE to avoid leaking FK state to other tests in the same connection. Refs: regression sweep 2026-08-06; pre-existing since 5d55146 (Phase 4).
The 203791d refactor moved Forge execution primitives into the kernel package. The executor now imports BudgetStatus and get_task_store from animus_kernel.{budget,db}, but the budget tests still patched the forge-side equivalents and instantiated the forge-side BudgetManager. Three classes of failure were caused by this drift: 1. test_executor_halts_on_effective_token_overspend instantiated forge BudgetManager whose BudgetStatus enum is distinct from kernel's. The executor's `if mgr.status == BudgetStatus.EXCEEDED` check failed and the executor fell through to can_allocate, producing "Token budget exceeded" instead of "Budget exceeded (effective-tokens)". Fix: import BudgetManager from animus_kernel.budget so the enum comparisons match. 2. test_daily_limit_blocks_when_exceeded and test_daily_sums_across_agents patched animus_forge.db.get_task_store, but the executor's runtime lookup is animus_kernel.db.get_task_store. The patch never landed. 3. test_daily_limit_blocks_after_threshold had the same patch target. Fix: relocate the patches to animus_kernel.db. The test's TaskStore instance remains from animus_forge.db but its get_daily_budget() interface is identical, so the kernel executor calls it without surprise. Refs: regression sweep 2026-08-06; pre-existing since 203791d.
… (ADL-20260806-001) The forge-side budget package was a byte-for-byte duplicate of animus_kernel.budget (plus forge-internal naming drift on imports). After the 203791d executor→kernel consolidation, two distinct BudgetManager classes and BudgetStatus enums existed side-by-side; tests using the forge-side class silently fell through the executor's kernel-side enum comparison (task #34, dd77401). Phase 1 of ADL-20260806-001 collapses the duplicate: every inner module becomes a 1-line 'from animus_kernel.budget.X import *' pass-through, and __init__.py re-exports the kernel surface with the same __all__. Verified: - animus_forge.budget.BudgetManager is animus_kernel.budget.BudgetManager - All 22 names in __all__ resolve to the kernel class (no fork) - 141/141 budget tests green - All ~30 forge-internal consumers resolve unchanged - Phase 2/3/4 (migrate imports, deprecation warning, removal) follow-up Co-Authored-By: Claude <noreply@anthropic.com>
…o animus_kernel.budget ADL-20260806-001 Phase 2. With Phase 1 in place, the forge-side animus_forge.budget package is a thin pass-through — every name in its __all__ resolves to the kernel class. Forge-internal consumers should import from the canonical kernel package directly, so there is exactly one source of truth and one place to find a class. 20 imports across 10 source files migrated: - agents/supervisor.py - analytics/collectors.py - api.py (2 imports) - api_routes/budgets.py (3 imports) - api_state.py - cli/commands/budget.py (3 imports) - cli/commands/consciousness.py - cli/helpers.py (3 imports) - coordination/consciousness_bridge.py - coordination/evolution_loop.py 14 mock.patch() test patches across 4 test files updated to point at the new import sites (animus_forge.budget.X is still bound to the same class via the pass-through, but tests must patch where the consumer reads from): - test_cli.py (5) - test_cli_coverage_boost.py (6) - test_collectors.py (2) - test_supervisor_process_message.py (1) - test_cli_consciousness_evolve.py (2) Verified: 307/307 tests pass across the touched test surface. Net diff: 35 lines in, 35 lines out — pure import-path migration. Phase 3 (DeprecationWarning on import animus_forge.budget) and Phase 4 (remove the package) follow-up. Co-Authored-By: Claude <noreply@anthropic.com>
…ge.budget (ADL-20260806-001)
ADL-20260806-001 Phase 3. The forge-side budget package is a thin
pass-through to animus_kernel.budget since Phase 1. Now that all
in-repo consumers (Phase 2) have migrated, any remaining external
consumer gets a hard signal that they should follow suit.
What you see:
- import animus_forge.budget → DeprecationWarning pointing at
animus_kernel.budget as the migration target
- Same warning on each of the 6 inner modules
- The ansible forge-side pyproject.toml promotes these to
'error' severity, so any consumer still importing animus_forge.budget
fails its OWN test suite with the migration message
Phase 2 migrated all 20 in-repo source imports to animus_kernel.budget.
Phase 3's deprecation now forces every repo-internal test that was
still importing through the forge-side package to migrate too — 17 test
files across test_budget_*, test_collectors, test_evolution_loop,
test_consciousness_bridge, test_executor_parallel, test_workflow_e2e,
test_cli_*, test_supervisor_*, test_c1_enforcement_loop, test_benchmarks.
Verified:
- 466/466 budget-relevant tests pass (the 11 failures + 8 errors in
test_c1_enforcement_loop, test_workflow_e2e, test_executor_parallel,
test_benchmarks are all pre-existing — independent of budget, see
REGRESSION_SWEEP_2026-08-06.md)
- animus_forge.budget import is now a hard DeprecationWarning that
pytest promotes to error → impossible to miss during consumer migration
- Identity + functional checks still pass: from animus_forge.budget
import BudgetManager is animus_kernel.budget.BudgetManager → True
Phase 4 (remove the package) follows.
Co-Authored-By: Claude <noreply@anthropic.com>
…260806-001)
ADL-20260806-001 closure. Phase 1+2+3 left the package as a thin
deprecation shim; every in-repo consumer was migrated to
animus_kernel.budget. The deprecation's hard-failure mode (Phase 3)
verified there were no remaining consumers — now the package is gone.
Removed:
- packages/forge/src/animus_forge/budget/ (7 files: __init__,
manager, models, persistence, preflight, strategies, cost_audit)
- error filterwarnings entries in pyproject.toml (the deprecation
is moot without the package)
Migrated (last consumer missed by Phase 2 grep):
- scripts/run_evolution.py imports BudgetConfig/BudgetManager from
animus_forge.budget.manager — moved to animus_kernel.budget.manager
Verified:
- 466/466 budget-relevant tests pass — identical pass/fail breakdown
to Phase 3 (the 11 failures + 8 errors are pre-existing, unrelated
to budget; see REGRESSION_SWEEP_2026-08-06.md)
- animus_forge.budget now raises ImportError — gone for good
- animus_kernel.budget remains the canonical source: BudgetManager
works, BudgetStatus enum is intact, all 22 names in __all__ resolve
ADL-20260806-001 closed: 4-phase migration complete. Single source of
truth for budget primitives.
Co-Authored-By: Claude <noreply@anthropic.com>
Trivial change to docs/getting-started/installation.md to match the workflow's paths filter. Activates the Pages deploy that has been ready since 2026-06-27 but blocked by an outdated memory entry that assumed the AreteDriver account was on Free plan. Verified 2026-08-07 via gh api /user: plan is Pro, which supports Pages for private repos. The workflow itself is unchanged — see ADL-20260808-001 in notes/decisions/2026-08.md for the full rationale and lesson.
…002) Replace the public-facing "exocortex" framing with "Mind-class AI operating environment" across PyPI surfaces, public docs, and package READMEs. Internal philosophical anchors (CLAUDE.md, Constitution, agent identity, architectural body where the metaphor is load-bearing) retain "exocortex" per the public/ private split now codified in BRANDING.md. Scope: - 41 modified: pyproject.toml (root + core/bootstrap/contracts), release/ package-matrix.yaml, 24 public doc files, 5 package READMEs, 2 architecture book intros (charter + overview with body philosophical-anchor references), PHASE3_INTELLIGENCE.md, README templates. - 2 new: BRANDING.md (public/private framing + decision rule) and scripts/ verify_exocortex_rebrand.py (deterministic regression contract). Verification: - scripts/verify_exocortex_rebrand.py → 5/5 PASS (PyPI clean, public docs clean, architecture intros reframed, 28 Bucket-B preservation zones retain 'exocortex', archive packages preserved). Exit 0. - packages/core/tests/test_cli_commands.py → 26 passed - packages/forge/tests/test_budget.py → 26 passed - All package imports verified: animus, animus_forge, animus_kernel, animus_bootstrap, convergent, animus_types.sensitivity. - 0 broken internal markdown links across changed docs. Decision (ADL-20260808-002, pending formal log): Bucket-B preservation zones require real body content. When an intro-reframed Bucket-D file's body never organically used "exocortex", the fix is to add a philosophical-anchor reference (charter.md, overview.md), not to weaken the verifier. Co-Authored-By: Claude <noreply@anthropic.com>
Three fixes applied per session-reviewer findings on f5f1658: 1. Phantom path: BUCKET_B_PRESERVE listed docs/CONSTITUTIONAL_PRINCIPLES.md (uppercase), but the file was relocated to docs/architecture/constitutional- principles.md in commit aae5be7. The verifier silently skipped the missing path, providing false confidence. Removed from the list; the philosophy is already anchored elsewhere in the preservation set (agent identity modules, consciousness-quorum bridge). 2. Missing-path guard: check_preservation_zones() now fails (instead of silently skipping) when a BUCKET_B path doesn't exist on disk. An attacker who deleted a preservation file would previously have caused the verifier to print "OK: 28 Bucket-B preservation zones retain 'exocortex'" while only verifying 27. With the guard, a missing path now produces a failure and the count stays honest. 3. Lint hygiene: ruff flagged F841 (unused 'failures' dict) and two F541 (f-string without placeholders) in the committed script. Now both clean. Verified: verifier still 5/5 PASS, ruff clean, missing-path guard catches phantom entries (synthetic test confirms it fails on a bogus path). Co-Authored-By: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
ADR-007 + ADR-008 — runtime lifecycle foundation
Accepts the runtime lifecycle architectural decisions and ships the
implementation, isolated test harness, and operator documentation.
See
docs/reviews/animus-runtime-lifecycle-evidence-packet.mdfor the full evidence packet. That document is the primary review
reference.
Scope
animus-runtime.targetas the single lifecycleboundary. Three deployment profiles
(
development-local,desktop-login,continuous-node). Four-stateProcessClassificationwith provenance rules.pgrepisforbidden in authoritative classification paths.
Commits (atomic, on
docs/adr-007-008)What this PR contains
docs/specifications/animus-runtime-lifecycle-build-spec.md).docs/specifications/animus-runtime-lifecycle-migration.md).(
packages/bootstrap/src/animus_bootstrap/lifecycle/).(
packages/bootstrap/tests/test_runtime_lifecycle/).docs/systemd/animus-runtime.md,docs/operations/process-registry.md).(
docs/reviews/animus-runtime-lifecycle-four-lens-review.md).docs/reviews/animus-runtime-lifecycle-evidence-packet.md).CLAUDE.mdupdates referencing the new package.6b92c7d, see Governor section below).Focused validation
cd packages/bootstrap PYTHONPATH=src pytest tests/test_runtime_lifecycle/ tests/test_runtime.py tests/test_runtime_e2e.py -vResult:
139 passed, 1 skipped(in ~12 s).The 1 skipped test is environment-dependent in the pre-existing
tests/test_runtime.py(AnimusRuntime orchestrator suite).Spec test matrix coverage
20 / 20 cases from the build spec §16 are implemented in
tests/test_runtime_lifecycle/. Coverage table in the evidence packet.Hard-constraint audit
maindocs/adr-007-008pgrepfrom authoritative classification pathsFakeSystemdonlyenable-lingeronlyAdversarial review
19 findings across architect, Linux/systemd, reliability, and
security/red-team lenses. 17 closed during the review (including
two code fixes: rollback
daemon-reloadwarning and expandedverification properties MemoryMax+KillMode+CPUQuota+Delegate). Two
tracked as Phase 7 spec followups (consent log path, drop-in
directory permissions).
Governor adapter (ADL-20260805-001, commit
6b92c7d)Wires the standalone
animus-loop-governorcontrol plane as averifier citizen in Forge's mission lifecycle. Implements the first
two of three ADL follow-ups; the third (real-
algintegration smoke)is gated on the wheel being installable in CI.
Headline property: The worker may claim completion. Only the
Governor decides completion.
Public API (one seam)
What changed
packages/forge/src/animus_forge/governor/(9 files,~1700 LOC including the protocol mirrors). Subprocess-only — never
import animus_loop_governor.*(vendor-independence preserved).packages/forge/tests/test_governor/(12 files,11 fixture JSONs). All subprocess calls bypassed via
FakeGovernorClienttest double; autouse_isolate_alg_pathstrips
PATHunlessANIMUS_LOOP_GOVERNOR_INTEGRATION=1.MissionScheduler(packages/forge/src/animus_forge/scheduler/mission_scheduler.py,+165/-1): adds
_start_ready_mission()that gatesREADY → RUNNINGonensure_run()success. Failure leavesmission in
READYfor the next tick (noBLOCKEDstatus in theenum). Persists
mission.metadata["governor_run"]atomicallywith the transition.
Strict compatibility
A known run id is only reused when the
CompatibilityKeymatchesexactly: same canonical repository path, same mission id, same
policy version, same adapter version. Mismatch →
RunUnusableError,not silent fallback. Cross-mission isolation: filesystem-hint Step 2
falls through silently on mission mismatch.
Exit-code contract
algrcGovernorError(sniffed:PermissionDenied/ContractIntegrity)FAILEDcompileContractRejectedErrorREADYverifyVerifyDeniedErrorneeds_repairGovernorErrorFAILEDRuntimeErrorVerification gates
cd packages/forge PYTHONPATH=src pytest tests/test_governor \ --cov=src/animus_forge/governor --cov-fail-under=97 -v ruff check src/animus_forge/governor tests/test_governoradapter.py100% covered; total 98.63% (≥97% gate)sk-ant-secret/sk-openai-secretto verify sanitized-envstripping)
test_missions.pyortest_citizens.py(161 pass,1 pre-existing SQLite FK-cascade flake deselected — unrelated)
Open items (do not block)
Attribution: the broader bootstrap suite has 42 failures
under full-suite ordering. They are characterized as existing
full-suite order-interaction failures not reproduced in the
focused lifecycle suite. An attribution comparison against
origin/mainis the appropriate followup to make the"pre-existing and unrelated" claim defensible. This evidence
packet does not prove that — the claim is only strongly
suggested by the focused-suite isolation behavior.
Spec followup:
consent_log_pathforcontinuous-nodeconsent evidence.
Spec followup: drop-in directory
chmod 700in theinstaller.
ADL-20260805-001 third follow-up: real-
algintegrationsmoke gated on
ANIMUS_LOOP_GOVERNOR_INTEGRATION=1. The wheelis vendored at
~/Downloads/animus_loop_governor-0.1.0-py3-none-any.whlbut not yet installable in CI. Friction Report template lives
at
~/projects/animus-loop-governor/.github/ISSUE_TEMPLATE/friction-report.md(the
animus-loop-governor/directory is not yet a git repo).Reviewer checklist
Read
docs/specifications/animus-runtime-lifecycle-build-spec.md§3-§11.Read
docs/reviews/animus-runtime-lifecycle-four-lens-review.md.Read the evidence packet (
docs/reviews/animus-runtime-lifecycle-evidence-packet.md).Read
packages/forge/src/animus_forge/governor/CLAUDE.mdforadapter constraints (vendor-independence, sanitized env,
strict compatibility).
Run the focused suite in a fresh shell:
git checkout docs/adr-007-008 cd packages/bootstrap PYTHONPATH=src pytest \ tests/test_runtime_lifecycle/ \ tests/test_runtime.py \ tests/test_runtime_e2e.py -vExpected:
139 passed, 1 skipped.Run the focused governor suite:
cd packages/forge PYTHONPATH=src pytest tests/test_governor \ --cov=src/animus_forge/governor --cov-fail-under=97 -vExpected:
112 passed, coverage ≥ 97%.(Optional) Run the attribution baseline against
origin/main:Compare the failure names, failure count, first failure, and
exit code against
/tmp/bootstrap-tests.logfrom this branch.🤖 Generated with Claude Code
Co-Authored-By: Claude noreply@anthropic.com