Repository navigation
feat: promote lab workflow improvements for 0.3.0 - #7
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: 4594582687
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".
| ...LEARN_INDEX_FILES.map(fileName => ({ | ||
| sourcePath: path.join('templates', '.omp-flow', 'learn', ...fileName.split('/')), | ||
| destinationPath: path.join('.omp-flow', 'learn', ...fileName.split('/')), | ||
| group: 'core' as const, | ||
| })), |
There was a problem hiding this comment.
When a user customizes any .omp-flow/learn/**/index.md and later runs omp-flow init --force or omp-flow update --force, registering these authored indexes as core managed resources causes them to be replaced by the packaged templates; update's protected user-data prefixes do not include Learn, and forced init provides no backup. Seed these files only when missing, as is done for the Wiki root, rather than treating durable Learn content as overwriteable templates.
Useful? React with 👍 / 👎.
| for path in candidates: | ||
| if path.is_file(): | ||
| return path |
There was a problem hiding this comment.
When an OMP or Claude installation customizes its harness-native omp-flow-sleep copy, _skill_path still returns the universally installed .agents copy first, so harvesterRevision and the deterministic run receipt describe a different Skill from the one the native actor loads. Changing the actual harvester then cannot create a revised run because the unchanged .agents hash reproduces the existing receipt; either bind the executing Harness explicitly or fail when installed copies diverge.
AGENTS.md reference: AGENTS.md:L87-L93
Useful? React with 👍 / 👎.
| untracked = _git(repo, "ls-files", "--others", "--exclude-standard", "--", source_relative) | ||
| if untracked: | ||
| raise WorkflowError("Task Bundle contains untracked files") |
There was a problem hiding this comment.
When a Task Bundle contains an ignored file such as *.log, git diff remains clean and git ls-files --others --exclude-standard deliberately omits that file, while _directory_digest still includes its bytes and the archive is reported as a ready Git checkpoint. Sleep can therefore derive Candidates from evidence absent from sourceTree, and a later clone or recovery of the claimed checkpoint silently loses that evidence; detect ignored files too or derive the archived digest exclusively from the recorded Git tree.
AGENTS.md reference: AGENTS.md:L101-L103
Useful? React with 👍 / 👎.
| run["assignment"] = _sleep_assignment(run) | ||
| atomic_write_json(_run_path(repo, receipt), run) | ||
| return {"run": run, "assignment": run["assignment"]} |
There was a problem hiding this comment.
start_sleep_run renders, persists, and returns an independent ompFlowSleep executable assignment, creating a second dispatcher outside operation start. That bypasses the common operation, predecessor, actor, and external-receipt correlation path and gives Harness integrations two incompatible dispatch and recovery protocols; route Sleep through the existing operation assignment producer rather than introducing a parallel one.
AGENTS.md reference: AGENTS.md:L70-L74
Useful? React with 👍 / 👎.
| atomic_write_text( | ||
| index, | ||
| "---\nokf_version: \"0.2\"\n---\n\n# Wiki Sleep\n\n" | ||
| "Candidate knowledge awaiting review and possible Wiki promotion belongs here.\n", | ||
| ) |
There was a problem hiding this comment.
On the first Sleep run, _ensure_sleep_root makes the Python kernel author a Git-visible Markdown index containing the purpose and placement semantics for Sleep knowledge. This manufactures authored project knowledge as a runtime side effect and leaves users with semantic content that is hard-coded outside the canonical template and Skill sources; bootstrap it through a create-if-missing installation or an assigned authoring Skill instead.
AGENTS.md reference: AGENTS.md:L8-L15
Useful? React with 👍 / 👎.
1 participant
大更新；增加了learn与sleep机制