Give the Kanon worker PR-template and testing conventions

[kelos-bot]

What type of PR is this?

/kind cleanup

What this PR does / why we need it:

Updates the kanon-workers agent's standing instructions (agentsMD) in
kanon-development/kanon-workers.yaml with two conventions learned from recent
Kanon (kelos-dev/kanon) PR reviews. The kanon-workers agent is the one that
opens and updates PRs against the kanon repo, which has no CLAUDE.md, so
project conventions must live in the agent config.

1. PR template / label gate. After Kanon PR #8 ("Add label workflow and PR
template", merged 2026-05-30), the kanon repo enforces a check-pr-labels CI
gate: a PR keeps the blocking needs-kind / needs-release-note labels (and
fails CI) unless its body carries a /kind line and a filled release-note
block. A worker PR opened after #8 would trip this gate. Added a
## Pull Requests section requiring the agent to follow
.github/PULL_REQUEST_TEMPLATE.md — mirroring how the kelos repo's own
CLAUDE.md supplies the same guidance for self-development/kelos-workers.

2. Testing conventions. The Kanon reviewer repeatedly flagged the same
test-coverage gaps across worker PRs:

• Kanon PR Add axon CLI for managing Tasks #9 (P3) — the new update command had no test for the pull→apply
  happy path, only the no-op "No changes." branch.
• Kanon PR Add config file support with direct token credentials #16 (P2) — the prune safety guard (a prunable file changed outside
  kanon becomes a conflict unless --adopt) was untested in both directions,
  and project-scope pruning was only exercised indirectly via agent scope.
• Kanon PR Rename Go module to gjkim42/axon and add CLI install instructions #11 (P3) — dropping permission rendering added no negative assertion
  that the rendered output no longer contains the permission keys, so a
  regression re-introducing them would not be caught.

The worker agentsMD had no testing guidance at all. Added a ## Testing
section telling the agent to: test the happy path / primary action (not only
early-return guards), assert both directions of a two-outcome guard, cover
flag/scope-narrowed behavior directly, and add a negative assertion when
removing behavior. This pushes the lesson to the point where the code is
written instead of relying on review to catch the same gaps each time.

Which issue(s) this PR is related to:

N/A

Special notes for your reviewer:

Scope is limited to kanon-development/kanon-workers.yaml. Both changes put the
convention where the kelos repo puts its equivalents — in the agent's standing
instructions — because the kanon repo has no CLAUDE.md to carry them.

Does this PR introduce a user-facing change?

NONE

[gjkim42]

Why do we need to refine kanon-development? self-development has the same instructions but it works fine.
/kelos pick-up

[kelos-bot]

🤖 Kelos Task Status

Task kelos-pr-responder-pull-request-review-658ffd796fb0 has succeeded. ✅

[kelos-bot]

🤖 Kelos Agent @gjkim42

Good question — I dug into why self-development works without this and the answer points to a real gap rather than a redundancy:

kelos-workers complies with the PR template not because of its prompt, but because the kelos repo has a CLAUDE.md ("Always follow .github/PULL_REQUEST_TEMPLATE.md…"). That standing guidance is what clears the label gate. The kanon repo has no CLAUDE.md (verified via the GitHub API and a code search), so kanon-workers got no equivalent guidance from anywhere — and after kanon PR #8 added check-pr-labels, its PRs would trip the gate.

I've reworked the change to match how self-development does it: the convention now lives in the kanon-workers-agent agentsMD (the agent's standing instructions, the kanon analog of kelos's CLAUDE.md), and the numbered procedure steps are reverted to be identical to kelos-workers. Net diff is now just a 10-line ## Pull Requests section.

If you'd prefer this live in a CLAUDE.md added to the kelos-dev/kanon repo itself (which would also cover the other kanon agents and human contributors), I'm happy to open that as a follow-up against kanon instead — let me know.