Repository navigation
feat: machine-safe automate-me - #5
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4398481383
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".
| Ask where the skill goes: | ||
|
|
||
| - **Repo** (`skills/<handle>-mode/SKILL.md`, or the existing mode skill's path in update mode): shared through git. Only for patterns mined from personal, non-confidential history. | ||
| - **Local layer** (`~/.agents/skills/<handle>-work/SKILL.md`): never committed. Use it for anything mined from work history. Copilot CLI, Codex, and others read `~/.agents/skills`. |
There was a problem hiding this comment.
Skip PR landing for the local destination
When the user selects this local work-history layer, step 6 still unconditionally instructs the agent to create a worktree, commit, and open a PR. Because this destination is outside the repository and is explicitly promised to be “never committed,” the flow either fails at landing or risks publishing the data it was meant to keep local; make the landing procedure conditional on the selected destination.
Useful? React with 👍 / 👎.
| Ask where the skill goes: | ||
|
|
||
| - **Repo** (`skills/<handle>-mode/SKILL.md`, or the existing mode skill's path in update mode): shared through git. Only for patterns mined from personal, non-confidential history. | ||
| - **Local layer** (`~/.agents/skills/<handle>-work/SKILL.md`): never committed. Use it for anything mined from work history. Copilot CLI, Codex, and others read `~/.agents/skills`. |
There was a problem hiding this comment.
Keep the local layer consistent with mode handling
When the local option is chosen, the generated directory and required frontmatter name are <handle>-work, but step 0 only discovers *-mode skills and line 85 still advertises /<handle>-mode. The local skill therefore cannot be found for repeat updates and its documented invocation does not match its name; either retain <handle>-mode for this destination or update discovery and trigger instructions consistently.
AGENTS.md reference: AGENTS.md:L14-L14
Useful? React with 👍 / 👎.
| - **Repo** (`skills/<handle>-mode/SKILL.md`, or the existing mode skill's path in update mode): shared through git. Only for patterns mined from personal, non-confidential history. | ||
| - **Local layer** (`~/.agents/skills/<handle>-work/SKILL.md`): never committed. Use it for anything mined from work history. Copilot CLI, Codex, and others read `~/.agents/skills`. | ||
|
|
||
| If the destination is inside a git repo with a remote (`git remote -v` prints something) and the mined scope included work history, stop and warn: "This will publish rules mined from work history to <remote>." Proceed only on an explicit yes. |
There was a problem hiding this comment.
Update the guide for the work-safe destination contract
This introduces a local-only destination and a confirmation gate before publishing work-derived rules, but docs/guide/09-make-it-yours.md:11 still says the skill always writes a Cursor-specific path and opens a PR. That leaves the user-facing instructions describing the unsafe, removed flow instead of the new contract, so update the guide together with this change.
AGENTS.md reference: AGENTS.md:L17-L17
Useful? React with 👍 / 👎.
automate-me now asks which history to mine (this workspace by default, the whole machine only on an explicit yes) and where the skill goes: the repo, or a local ~/.agents/skills layer that is never committed. It warns before publishing rules mined from work history to a remote. Per-harness transcript paths live in one shared reference.
Verified with a headless Claude Code dry run. Stacked on the bridge PR.
🤖 Generated with Claude Code