[entelecheia]

Refs #40
Refs #13

The README opens with install, setup preview/person apply, candidate review and session-end/manual ingest cadence.

CI's Ubuntu/macOS matrix gains an installed-artifact smoke after the unchanged budget steps. It packs this checkout, installs the local tarball offline under a fresh HOME, and runs the installed .bin/amem alias outside the checkout. Public-API fixture seeding also resolves inside that artifact.

The smoke verifies:

• version/help and all-target setup preview, including the installed entry in registration commands;
• a complete before/after HOME snapshot (path/type/size/mode, regular-file SHA-256 and symlink target), plus six instruction/rule sentinels and native-client traps;
• MCP initialize, tools/list descriptions and exactly one seeded curated search hit; every captured message rejects the legacy program name, including underscore forms;
• quiet PostToolUse output, exactly one newly appended hook metric, AND this run's distinct event marker in a capture row or a committed root/capacity hook-spool record. Metric-only fail-open outcomes refuse.

Child config/cache/tmp/home paths are isolated; credentials and agent authority variables are not inherited. The smoke disables only the Bun runtime transpiler cache, whose .pile files otherwise changed HOME during setup module loading. The whole HOME comparison has no excluded paths. Existing R2/R3 environments, thresholds, counts and advisory policy remain unchanged. Owned command groups use a cached three-second reap deadline and bounded cleanup.

Validation at 7e59115, based on main 8350589:

• Frozen install, TypeScript and boundary pass: boundary: OK (71 source files).
• Guards/argv checks: 17 pass, 0 fail (116 assertions).
• Actual macOS offline packed-install smoke passes with durable hook evidence and unchanged HOME.
• Two actual installed-copy faults refuse at their intended guards: throwing capture staging despite its fail-open metric, and an undeclared setup HOME write. Both private HOME trees are removed. Checkout product files are unchanged; the script-only afterInstall fixture seam is not exposed through CLI/env.
• Full suite: 1258 pass, 2 skip, 0 fail; 1260 tests across 130 files (211.08s), plus npm pack dry-run (79 files).
• Native hooks R2 passes: max workload p95 48.24ms / peak 79.94MB; timeout probes pass, zero stalled attempts/retained processes.
• Native MCP R3 passes: idle 47.83MB / peak 176.70MB, correct corpus/results, EOF child reaped.
• Fresh-main diff check lists only the four delivered files; OCR/diff checks completed.

DGX SSH was unreachable, so a local Linux leg did not execute. Ubuntu/macOS CI remain required before merge, including AC11. No real-home configuration mutation, account/model turn or publication was performed.

Approved-review nit delta at b9b8df1:

• Snapshot capture ids and normalized published-spool identities before the hook; only new records qualify, including across capacity/draining renames. Unknown snapshots refuse.
• Check the installed entry in every client hook command.
• Preserve the first command error if cleanup also fails; cleanup-only failures remain hard.
• Frozen install, TypeScript, boundary, actual installed smoke and both installed-copy fault probes pass. Focused guards: 19 pass, 0 fail, 131 assertions. Full suite: 1260 pass, 2 skip, 0 fail (1262 tests, 130 files, 208.00s); pack dry-run: 79 files.
• R2/R3 measurements above belong to 7e59115; this script/test-only delta does not change product or budget code. Exact-head Ubuntu/macOS CI remains required.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-24T13:37:17.684842Z	bb1388e	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.