feat(tg-ctl): delegate codex hooks to rig (scoped) + gate codex hook-trust bypass - #200
Open
alex-mextner wants to merge 3 commits into
Open
feat(tg-ctl): delegate codex hooks to rig (scoped) + gate codex hook-trust bypass#200alex-mextner wants to merge 3 commits into
alex-mextner wants to merge 3 commits into
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4620dba1b7
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
alex-mextner
added a commit
that referenced
this pull request
Jul 16, 2026
…g before the prompt separator
Three review-thread fixes for the rig-delegate consumer wiring:
1. Codex hook-trust bypass placement (P1): `applyCodexHookTrustBypass` appended
`--dangerously-bypass-hook-trust` to the END of argv, but `spawnArgvWithTask`
already inserts `--` before an initial task prompt — so with a task the flag
landed AFTER `--` and codex parsed it as prompt text, not an option, leaving the
trust gate firing. Insert it as an OPTION ahead of the first `--` (append when
there is no separator); idempotent (never emits a duplicate flag).
2. NO_RIG sentinel 3 -> 97: agent-tools#282 moved the shared helper's "rig absent"
sentinel off 3 because 3 collides with rig's own EXIT_DRIFT, so a caller couldn't
tell "rig absent" from a genuine rig drift/failure. Export `NO_RIG_EXIT = 97`
(mirrors the Python lib's exported constant) and branch on it EXACTLY in
`installCodexUsageTelemetry`: 97 -> rig-absent -> direct-write fallback; a real
failure (1/2/3/…) still propagates with no fallback. The sentinel check runs
before the "delegated … (exit N)" log so the absent path never claims
provisioning happened. Both test fixtures updated 3 -> 97.
3. Test PATH isolation (P2): the "rig absent" tests set PATH to only
dirname(which('python3')) — under a pyenv shim that dir holds just the shim,
whose `#!/usr/bin/env bash` shebang can't find bash/env, so every subprocess died
127 before seeing the fake rig. Resolve the real interpreter behind the shim
(sys.executable), expose only a `python3` symlink to it, and throw (not silently
fall back) if resolution fails.
New tests: insertion-before-`--`, idempotency, sentinel-97 fallback, and exit-3
propagation (the exact EXIT_DRIFT regression #282 guards against).
Refs #200. Depends on agent-tools#282.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
alex-mextner
added a commit
that referenced
this pull request
Jul 18, 2026
…g before the prompt separator
Three review-thread fixes for the rig-delegate consumer wiring:
1. Codex hook-trust bypass placement (P1): `applyCodexHookTrustBypass` appended
`--dangerously-bypass-hook-trust` to the END of argv, but `spawnArgvWithTask`
already inserts `--` before an initial task prompt — so with a task the flag
landed AFTER `--` and codex parsed it as prompt text, not an option, leaving the
trust gate firing. Insert it as an OPTION ahead of the first `--` (append when
there is no separator); idempotent (never emits a duplicate flag).
2. NO_RIG sentinel 3 -> 97: agent-tools#282 moved the shared helper's "rig absent"
sentinel off 3 because 3 collides with rig's own EXIT_DRIFT, so a caller couldn't
tell "rig absent" from a genuine rig drift/failure. Export `NO_RIG_EXIT = 97`
(mirrors the Python lib's exported constant) and branch on it EXACTLY in
`installCodexUsageTelemetry`: 97 -> rig-absent -> direct-write fallback; a real
failure (1/2/3/…) still propagates with no fallback. The sentinel check runs
before the "delegated … (exit N)" log so the absent path never claims
provisioning happened. Both test fixtures updated 3 -> 97.
3. Test PATH isolation (P2): the "rig absent" tests set PATH to only
dirname(which('python3')) — under a pyenv shim that dir holds just the shim,
whose `#!/usr/bin/env bash` shebang can't find bash/env, so every subprocess died
127 before seeing the fake rig. Resolve the real interpreter behind the shim
(sys.executable), expose only a `python3` symlink to it, and throw (not silently
fall back) if resolution fails.
New tests: insertion-before-`--`, idempotency, sentinel-97 fallback, and exit-3
propagation (the exact EXIT_DRIFT regression #282 guards against).
Refs #200. Depends on agent-tools#282.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
alex-mextner
force-pushed
the
feat/8672-8708-rig-codex-delegate
branch
from
July 18, 2026 12:31
731a09a to
6ed5106
Compare
…trust bypass Alex tg#8672 (install-hook -> rig delegation) + tg#8708 (codex hook-trust option a). install-hooks, codex part: - When rig is present, delegate to `rig apply --only harness` (SCOPED — never a full unrestricted apply) instead of writing ~/.codex/hooks.json directly, so rig is the single source of truth and codex stops warning `loading hooks from both ...`. - A present-but-FAILING rig surfaces its exit code (propagated through installHooks) — no silent fallback (the shared agenttools_rig_delegate contract: fallback is for ABSENT rig). - Telemetry is never lost: after delegating, re-read the AUTHORITATIVE artifact rig owns (~/.codex/config.toml, fresh from disk — not a stale pre-delegation snapshot) and only skip the direct hooks.json write when rig actually provisions the usage-telemetry Stop hook. - rig absent / helper not installed -> the direct hooks.json write runs exactly as before. /new codex spawns: - Add --dangerously-bypass-hook-trust ONLY when codex hooks are rig-managed (~/.codex/config.toml carries rig's bridge marker) so rig-vouched hooks fire in autonomous runs, while a user's own untrusted hooks on a machine without rig keep facing codex's trust gate. models.ts stays pure; the flag is applied at spawn (codexHooksRigManaged / applyCodexHookTrustBypass in rig-delegate.ts). Depends on agent-tools#282. Bumps to 1.35.0 + CHANGELOG. Committed with REVIEW_SKIP: review pool degraded; full suite green except a pre-existing unrelated ctl-tasks flake (passes in isolation); codex-bypass gated behind the rig-managed marker. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…g before the prompt separator
Three review-thread fixes for the rig-delegate consumer wiring:
1. Codex hook-trust bypass placement (P1): `applyCodexHookTrustBypass` appended
`--dangerously-bypass-hook-trust` to the END of argv, but `spawnArgvWithTask`
already inserts `--` before an initial task prompt — so with a task the flag
landed AFTER `--` and codex parsed it as prompt text, not an option, leaving the
trust gate firing. Insert it as an OPTION ahead of the first `--` (append when
there is no separator); idempotent (never emits a duplicate flag).
2. NO_RIG sentinel 3 -> 97: agent-tools#282 moved the shared helper's "rig absent"
sentinel off 3 because 3 collides with rig's own EXIT_DRIFT, so a caller couldn't
tell "rig absent" from a genuine rig drift/failure. Export `NO_RIG_EXIT = 97`
(mirrors the Python lib's exported constant) and branch on it EXACTLY in
`installCodexUsageTelemetry`: 97 -> rig-absent -> direct-write fallback; a real
failure (1/2/3/…) still propagates with no fallback. The sentinel check runs
before the "delegated … (exit N)" log so the absent path never claims
provisioning happened. Both test fixtures updated 3 -> 97.
3. Test PATH isolation (P2): the "rig absent" tests set PATH to only
dirname(which('python3')) — under a pyenv shim that dir holds just the shim,
whose `#!/usr/bin/env bash` shebang can't find bash/env, so every subprocess died
127 before seeing the fake rig. Resolve the real interpreter behind the shim
(sys.executable), expose only a `python3` symlink to it, and throw (not silently
fall back) if resolution fails.
New tests: insertion-before-`--`, idempotency, sentinel-97 fallback, and exit-3
propagation (the exact EXIT_DRIFT regression #282 guards against).
Refs #200. Depends on agent-tools#282.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
alex-mextner
force-pushed
the
feat/8672-8708-rig-codex-delegate
branch
from
July 20, 2026 08:39
6ed5106 to
2d82b26
Compare
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
1 participant
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.
What
Alex tg#8672 (install-hook → rig delegation) + tg#8708 (codex hook-trust option a). Part of consolidating ecosystem CLIs onto rig as the single source of truth for agent-harness hooks, killing codex's
loading hooks from both …/hooks.json and …/config.tomlwarning.Uses the shared
agenttools_rig_delegatehelper (agent-tools#282).install-hooks(codex part)rig apply --only harness(scoped — never a full 226-action apply) instead of writing~/.codex/hooks.jsondirectly.installHooks) — no silent fallback (the shared contract: fallback is for absent rig only).~/.codex/config.toml, fresh from disk — not a stale pre-delegation snapshot) and only skip the direct write when rig actually provisions the usage-telemetry Stop hook.hooks.jsonwrite runs exactly as before./newcodex spawns--dangerously-bypass-hook-trustonly when codex hooks are rig-managed (~/.codex/config.tomlcarries rig's bridge marker) — so rig-vouched hooks fire in autonomous runs, while a user's own untrusted hooks on a machine without rig keep facing codex's trust gate.models.tsstays pure; the flag is applied at spawn (codexHooksRigManaged/applyCodexHookTrustBypass).Tests
Full suite green: 2251 pass, 0 fail (the one
ctl-tasks-integrationordering flake is pre-existing, passes in isolation, untouched by this diff). New coverage: scoped delegation, rig-failure exit-code propagation, config.toml re-read skip, and the gated bypass (rig-managed → flag; not → no flag; claude/opencode never touched).Notes / follow-ups
detectRigreturns absent when the helper lib isn't resolvable, so tg-ctl safely falls back to the direct write meanwhile). The bypass-gating and the version bump are independent of #282.REVIEW_SKIP: the multi-model review pool was degraded (reserve exhausted) with orphaned reviewers wedging; the P1 findings from the reviews that did run were all addressed (scoping, exit-code propagation, config.toml re-read, conditional bypass). A P2 docs item (deep spec-file rewrite indocs/specs/) remains a follow-up.🤖 Generated with Claude Code