update default context - #223
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: 3530ff944e
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| "\n" + | ||
| "Keep standing memory small. If an active commitment or unresolved task must remain visible, create `~/context.d/20-open-loops.md` and remove it once resolved.\n" + | ||
| "\n" + | ||
| "Read `skills show memory` for memory workflows.\n"; |
There was a problem hiding this comment.
On an existing deployment upgraded without rerunning sys.bootstrap, ensurePersonalAgent replaces the old self-contained memory instructions with this redirect, but memory and process-orchestration are copied into a user's skills.d only by seedRepoSkillsToHome during setup/bootstrap. Consequently, skills show memory and the new system-level skills show process-orchestration instruction fail for upgraded users, leaving them without the workflows that were removed from standing context. Seed these new skills as part of the upgrade/reconciliation path, or retain the inline guidance until the referenced skill is present.
AGENTS.md reference: AGENTS.md:L46-L46
Useful? React with 👍 / 👎.
3530ff9 to
becee56
Compare
becee56 to
1abd184
Compare
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1abd1841fb
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| const config = ctx.config; | ||
| const uid = ctx.identity?.process.uid ?? 0; | ||
| const owner = resolveOwnerIdentity(ctx); | ||
| const builtinSkillsReady = ensureBuiltinSkillsForPrompt(ctx, owner); |
There was a problem hiding this comment.
When another process creates a skill at a newly bundled path after reconciliation reads it as missing but before ripgit.apply, this new prompt-time call can overwrite that user-authored file: the helper emits an unconditional put, and RipgitClient.apply is invoked without expectedHead, so it applies against the latest head rather than the state that was inspected. Use an atomic create-if-absent operation or bind the apply to the observed head and retry after rereading.
AGENTS.md reference: AGENTS.md:L11-L11
Useful? React with 👍 / 👎.
| const reconciledPaths = parseBuiltinSkillManifest(manifestResult); | ||
| const pendingSkills = BUILTIN_SKILL_FILES.filter((skill) => !reconciledPaths.has(skill.path)); |
There was a problem hiding this comment.
On every home created by the previous release, the manifest is absent, so reconciledPaths starts empty and all bundled paths become pending. If a user previously deleted one of the already-offered built-ins such as browser-target, gsv-manual, or skill-authoring, the first prompt after upgrading recreates it; the manifest only protects deletions made after this rollout. Initialize legacy homes with the prior catalog marked as reconciled while leaving only genuinely new paths eligible.
AGENTS.md reference: AGENTS.md:L11-L11
Useful? React with 👍 / 👎.
| "\n" + | ||
| "Keep standing memory small. If an active commitment or unresolved task must remain visible, create `~/context.d/20-open-loops.md` and remove it once resolved.\n" + | ||
| "\n" + | ||
| "Read `skills show memory` for memory workflows.\n"; |
There was a problem hiding this comment.
For a process running as a distinct agent whose home already contains an agent-specific skill named memory, the newly seeded owner-side built-in produces both user:memory and agent:memory. resolveSkillDocument deliberately rejects the unqualified name when multiple documents share it, so this standing instruction makes skills show memory fail as ambiguous in a supported layered-skill configuration. Refer to the qualified ID from the rendered skill index or change resolution to select the intended layer.
Useful? React with 👍 / 👎.
1abd184 to
96e28c7
Compare
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 96e28c72b5
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| DELETE FROM config_kv | ||
| WHERE key = 'config/ai/context.d/10-runtime.md' |
There was a problem hiding this comment.
When an installation already has an explicit config/ai/context.d/00-runtime.md entry, the preceding INSERT OR IGNORE skips the rename and this unconditional delete permanently discards the distinct 10-runtime.md override; the repeated 00-gsv.md/01-gsv.md pair has the same problem. Because context.d accepts arbitrary operator-managed filenames, preserve both values under a non-colliding key or otherwise migrate them without silently deleting prompt configuration.
AGENTS.md reference: AGENTS.md:L102-L102
Useful? React with 👍 / 👎.
| processOverrides, | ||
| ); | ||
| const timezone = config.get("config/server/timezone") ?? "UTC"; | ||
| await builtinSkillsReady; |
There was a problem hiding this comment.
For every RIPGIT-backed model run, resolveAiConfig() now blocks here on seven remote readPath calls even after all built-ins have already been seeded; with indexing enabled, collectPromptSkillIndex() then rereads the same tree and skill files, and the reconciliation still runs when indexing is off. This adds avoidable storage latency to every generation rather than only upgrades or missing-skill cases, so cache/reconcile at provisioning or combine it with the index traversal.
AGENTS.md reference: AGENTS.md:L47-L48
Useful? React with 👍 / 👎.
1 participant
No description provided.