Skip to content

Docs/realign worker docs with protocol 9 - #168

Merged
rafael-adcp merged 2 commits into
mainfrom
docs/realign-worker-docs-with-protocol-9
Aug 2, 2026
Merged

Docs/realign worker docs with protocol 9#168
rafael-adcp merged 2 commits into
mainfrom
docs/realign-worker-docs-with-protocol-9

Conversation

@rafael-adcp

Copy link
Copy Markdown
Owner

No description provided.

rafael-adcp and others added 2 commits August 1, 2026 20:49
Four statements drifted from the contract as protocol/8 and /9 landed. None
changes behaviour; each is a place where a doc still describes a retired design.

- SKILL.md's inline authorization block named only refs/kraken/claims/, omitting
  the state record protocol/9 made every terminal transition write. That block is
  the one copy that must stand on its own — SKILL.md hands it verbatim to a
  subagent precisely because that subagent never reads PROTOCOL.md — so it
  under-authorized the writes the skill goes on to order.

- DELIVERY.md still described the requeue gesture as "comment and remove
  awaiting-merge". protocol/8 dropped the asymmetry and /9 §3.1 makes removing a
  held label by hand a no-op on a task that has a record: the comment alone
  requeues. DELIVERY.md had not been touched since before either revision.

- PROTOCOL.md §12 documented four exit codes while the reference implementation
  also returns 3 (nothing startable) and 13 (unknown project). A driver written
  to §12 — which is what §12 exists to enable — reads an `idle` envelope as an
  unknown failure. Name both and fence them off as ergonomics, which §12 already
  allows above the wire contract.

- AGENTS.md summarized PROTOCOL.md without the state record, protocol/9's
  headline. The file's version marker was updated in #166; this sentence was not.

Also drop §3's label-derived reading of `queued`: since /9 the record decides,
and `kraken-task` is queue membership rather than a state — the same section goes
on to forbid the reading its own table implied.

Prose only. No wire-contract change, so no PROTOCOL_VERSION bump and no
HISTORY.md entry: these are clarifications, which the versioning rule amends in
place.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Lift the "Doesn't this already exist" heading and entry above "Why not just use
X?", so the section lands inside that entry's disclosure: the question a
first-time reader arrives with is what is visible, and expanding it reveals the
side-by-side that answers it rather than leaving the two to reference each other
across the page.

Also fix the "bellow" typo in the entry's prose.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@rafael-adcp
rafael-adcp merged commit c5b9b02 into main Aug 2, 2026
1 check passed
@rafael-adcp
rafael-adcp deleted the docs/realign-worker-docs-with-protocol-9 branch August 2, 2026 00:29
rafael-adcp added a commit that referenced this pull request Aug 2, 2026
#168 is the squash-merge of this branch's own two docs commits, so it carries no
content this branch does not already have — the merge is an ancestry link and the
resulting tree is byte-identical to the commit before it.

One conflict, in DELIVERY.md's "The branch" paragraph, because both sides edited
the same lines relative to the merge base and the squash left no ancestry to
match them up:

- main  — fixed the requeue gesture ("that comment alone requeues", dropping the
  retired "and removing awaiting-merge"), keeping the paragraph that has the
  worker rediscover its earlier PR from `brief.body` and the thread.
- ours  — the same gesture fix, plus c1: that rediscovery is replaced by reading
  `pr` off the execute envelope, which is where the program now puts it.

Resolved as ours: it is a strict superset — the fix main brought is already in
it, and reverting to the rediscovery paragraph would undo c1 in the one file that
tells a worker where to push.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant