Conversation
e319e3a to
82d1d54
Compare
9ea0d33 to
ec80a84
Compare
ucirello
changed the title
027dbf5 to
ffa7e62
Compare
ucirello
changed the title
25415d8 to
f2548d6
Compare
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
ucirello
changed the title
f2548d6 to
c0beb1f
Compare
![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: f2548d6c3e
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".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 533ca69f33
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".
20d75bb to
2c3e6b5
Compare
ucirello
changed the title
2c3e6b5 to
f6c5a90
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f6c5a90c37
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".
f6c5a90 to
3f9a4c9
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3f9a4c9e5e
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".
3f9a4c9 to
eda82f3
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: eda82f3fb6
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".
eda82f3 to
df491e5
Compare
You can now tell CE how hard the model should think when it writes code, when it writes a plan or brainstorm, and when Codex reviews. If you don't set it, nothing changes (still high). If you pick a value a tool can't use, we skip it. We never quietly pick a different one. You can also say this in an LFG run, including OpenCode. If you don't say how hard, we use your checkout config, or high. OpenCode gets this as --variant. Codex review now also accepts max and none. Fixes #1569 Fixes #1415 Fixes #1565
df491e5 to
b007d37
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b007d375c1
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".
| - `work_engine_preferences`: one or more ordered candidate objects | ||
| - `harness`: `codex | claude | grok | cursor | opencode` | ||
| - optional `model`: a model id or family understood by that harness; omission means its configured default | ||
| - optional `effort`: default high; a token the route cannot honor is omitted, never substituted |
There was a problem hiding this comment.
When a preference supplies a valid shared token that the selected route cannot honor, such as minimal for Claude, this rule only says to omit it. The authorization then records no effective effort, the adapter runs at its default high, and the returned binding still echoes the requested token without any instruction to report that it was dropped. This makes an explicit pin appear honored and misses the added plan's R5 requirement for a named reason; require the routing layer to disclose every omitted non-null effort pin while retaining the selected candidate.
AGENTS.md reference: AGENTS.md:L135-L137
Useful? React with 👍 / 👎.
|
|
||
| 1. **Native in-harness dispatch.** Attempt the platform subagent primitive with a per-agent model override (e.g. `model: "fable"` on the Claude Code `Agent`/`Task` tool). Capability is proven by attempt, not self-assessment — a harness that can serve the model natively does; one that cannot fails the attempt and falls through. **Receipt rule (R6):** a native run whose serving-side receipt names a *different* model family than requested falls through to the next adapter; a run with *no* receipt proceeds and is recorded as unverified (it does NOT fall through). | ||
| 2. **Claude CLI.** Run the bundled `scripts/elevation-dispatch.sh` worker as a detached job (see Off-host dispatch). Available when `claude` is on PATH. Do not preflight authentication in the host command context: the detached worker's provider-capable call is authoritative, and an authentication failure there follows Recovery. | ||
| 2. **Claude CLI.** Run the bundled `scripts/elevation-dispatch.sh` worker as a detached job (see Off-host dispatch). Available when `claude` is on PATH. Read `plan_effort` / `brainstorm_effort` with the same ordinary-key rule as the model key and export `CE_ELEVATION_EFFORT_OVERRIDE` on `start` when the token is honored. Do not preflight authentication in the host command context: the detached worker's provider-capable call is authoritative, and an authentication failure there follows Recovery. |
There was a problem hiding this comment.
When plan_effort or brainstorm_effort contains a shared token that Claude cannot honor, such as minimal, this condition tells the orchestrator not to export it but never requires disclosure. The worker consequently falls back to high, while the Transparency section reports only that elevation fired, so users cannot tell that their explicit effort choice was ignored. State the condition that an unhonored non-null effort pin must be named as omitted before continuing with the default tier.
AGENTS.md reference: AGENTS.md:L135-L137
Useful? React with 👍 / 👎.
1 participant
You can now tell CE how hard the model should think when it writes code, when it writes a plan or brainstorm, and when Codex reviews.
If you don't set it, nothing changes (still high). If you pick a value a tool can't use, we skip it. We never quietly pick a different one.
You can also say this in an LFG run, including OpenCode. If you don't say how hard, we use your checkout config, or high.
OpenCode gets this as
--variant. Codex review now also acceptsmaxandnone.Fixes #1569
Fixes #1415
Fixes #1565
Not this PR:
maxandnone, which the Codex CLI accepts #1565 only)Security Disclosure
We check the effort value before handing it to a CLI. Bad values fail with a clear error. No secrets or permission changes.
Agent Disclosure