Repository navigation
Add the knowledge publisher (kb publish) for the GPU box - #210
Conversation
The only GitHub writer for knowledge PRs. It runs as the repository owner through the box's existing gh login, which never reaches the bot host or any model environment. - publisher: each round reads the service's signed control record over SSH (or a local directory) and does nothing when it is stale. Every outbox item goes through check_item (signature, expiry, generation, pause, and auto_merge in both the control record and the publisher's OWN adapter configuration). The publisher performs the item, records the result locally, then writes a kb-ack signed with the publisher key. Items it cannot verify or must not execute are neither performed nor answered. Every decision is traced to traces/publisher.jsonl. - idempotent: an item already performed only has its ack re-sent; completion records are written atomically, and a corrupt one is set aside for people, never guessed. open_pr reuses an open PR on its branch only when its head is exactly the rebuilt commit (and a draft for companions). It rebuilds the commit deterministically from the signed contents on base_sha in a scratch index, and refuses to overwrite an existing branch with other content. It also refuses any path that is not a governed knowledge page. - actions: open_pr and open_companion_pr (always a draft). post_verdict, enqueue and update_branch first check that the PR is open at the signed head; enqueue marks a paused draft ready, then runs gh pr merge --auto --match-head-commit. pause runs procedure P: dequeue, disable auto-merge, convert to draft (verified), then label kb:hold. close closes the PR. A post_verdict (issued only for an unpaused repository under its current generation) first removes kb:hold and confirms it is gone before commenting, so pause, then resume, then re-gate, then enqueue completes. - a failed pause or close is never acknowledged: it stays in the outbox and is retried every round until it succeeds. Malformed control records and items are traced and skipped; they never stop the daemon. - transport failures (ssh down while listing or delivering an ack) are traced, never fatal: the completion record stays and the ack is re-sent next round without repeating the action. - double gate: without ALLOW_POST=1 a round only records what it would do; pushing a branch also needs ALLOW_PUSH=1. - CLI: kb publish (--remote host:/state | --local dir) [--once]. It needs KB_SERVICE_PUBKEY, KB_PUBLISHER_KEY and KB_PUBLISHER_GIT_AUTHOR, and never opens the service ledger. 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. |
![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: 13ef822719
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".
| item = check_item(envelope, control, self.service_public_key, now=self.clock(), | ||
| repo_publishes=publishes, repo_auto_merge=auto_merge) |
There was a problem hiding this comment.
When a valid signed envelope is copied or renamed to another filename matching ITEM_ID, the publisher verifies the envelope but never checks that item.id == item_id. The completion lookup therefore uses the alias while _record_done uses the signed ID, leaving the alias perpetually uncompleted and re-executing it every round; a duplicated post_verdict item, for example, posts the same gate comment indefinitely. Reject filename/payload ID mismatches before performing the action.
Useful? React with 👍 / 👎.
| for item_id in item_ids: | ||
| done = self._done_path(item_id) | ||
| if done.is_file(): |
There was a problem hiding this comment.
If two publisher processes share the configured state directory—for example, the daemon overlaps a manual --once invocation—both can observe that done is absent and execute the same item concurrently. Besides duplicate verdict comments, concurrent open_pr handling can let one process create the PR while the other records and sends a failed acknowledgement after losing the branch-push race, potentially moving the changeset to failed. Hold an interprocess lease across the completion check, action, and record write.
Useful? React with 👍 / 👎.
| self.github.gh("pr", "comment", number, "--repo", self.github.repository, | ||
| "--body-file", "-", input=str(item.body["comment"])) |
There was a problem hiding this comment.
If the process is killed after gh pr comment succeeds but before _record_done runs, the next invocation finds no completion record and posts the verdict again. Thus the advertised crash-safe behavior only holds for open_pr, while this action can create duplicate comments and trigger the gate workflow repeatedly. Before posting, adopt an existing comment carrying the exact signed verdict or otherwise persist a recoverable operation identity.
Useful? React with 👍 / 👎.
| args = ["pr", "create", "--repo", self.github.repository, "--base", "main", "--head", branch, | ||
| "--title", str(body["title"]), "--body-file", "-"] | ||
| self.github.gh(*args, *(["--draft"] if draft else []), input=str(body.get("body") or "")) |
There was a problem hiding this comment.
If kb pause runs after gh pr create succeeds but before the service collects this acknowledgement, the ledger still has the changeset in pr_requested, so pause_open_prs cannot see its PR number and issues no pause item. The later successful acknowledgement moves it to pr_open without checking the current pause state, leaving the newly created PR non-draft for the duration of the pause. The acknowledgement path needs to detect this state and immediately issue/apply procedure P.
Useful? React with 👍 / 👎.
| if done.is_file(): | ||
| try: | ||
| record = json.loads(done.read_text(encoding="utf-8")) | ||
| except ValueError: |
There was a problem hiding this comment.
A completion file containing valid JSON but not the expected acknowledgement object, such as {} or null, passes json.loads and is then handed to _send_ack, where indexing ack["item_id"] raises an uncaught KeyError or TypeError. Since serve does not catch failures from run_once, one such damaged or obsolete record terminates the publisher on every restart instead of being moved to .corrupt as intended. Validate the decoded record's type and required fields before resending it.
Useful? React with 👍 / 👎.
PR 7 of the knowledge-service redesign: the publisher (
kb publish). It runs on the GPU box and is the only GitHub writer for knowledge PRs, using the owner's existingghlogin.check_item(signature, expiry, generation, pause,auto_mergein both the control record and the publisher's own adapter config);kb-ack.pause/closeactions are retried every round and never acked as failed;open_pr: rebuilt from the signed contents onbase_sha; governed pages only; never overwrites a branch;open_companion_pr: always a draft;post_verdict: removeskb:holdand confirms it is gone first, so pause → resume → re-gate → enqueue completes;enqueue: ready, thenmerge --auto --match-head-commit;pause: procedure P;update_branch,close.ALLOW_POST=1, plusALLOW_PUSH=1for branch pushes; otherwise it only records.Commit hook approved; full suite and SPEC freshness pass locally.
🤖 Generated with Claude Code