cat-mode: Invoker plans set executionAgent claude explicitly

[EdbertChan]

Summary

Invoker runs plans by handing each step to an AI agent. Each step can name which agent does the work.

When a plan leaves that blank, a default agent is picked for you. That is not the one you want.

This change adds one rule. Every plan the assistant writes names Claude as the agent on each step.

It never relies on the default and never picks the other agent.

Two small example files and a test check the rule.

Review Claim

Every plan the assistant writes names Claude as the agent for each step.

Review Lane

behavior

Review Unit

corpus-lesson

Safety Invariant

Skill text plus small test files that only read that text. No hook or script changes.

Slice Rationale

Each standing rule is one reviewable claim, stacked in order. This one sits on top of the local-only rule in the same file.

Non-goals

• No hook or script changes.
• No change to Invoker itself.

Measured

Command: python3 -m unittest discover -s corpus/skills/cat-mode/tests

• base: Ran 3 tests in 0.007s -- OK (parent commit)
• head: Ran 4 tests in 0.019s -- OK

Test Plan

Test Plan

• python3 -m unittest discover -s corpus/skills/cat-mode/tests
• python3 scripts/ci/check_no_dated_provenance.py (ok, exit 0)
• python3 scripts/ci/check_skill_file_refs.py (ok, exit 0)
• python3 scripts/ci/check_skill_test_coverage.py --base origin/main --head HEAD (ok, exit 0)
• python3 scripts/ci/check_skill_test_debt_no_growth.py (ok, exit 0)

Revert Plan

Revert Plan

• Safe to revert? Yes
• Revert command: git revert <sha>
• Post-revert steps: None
• Data migration? No

🤖 Generated with Claude Code

---

Note

Low Risk
Corpus skill wording and read-only lesson tests only; no runtime, hook, or Invoker code paths change.

Overview
Adds a cat-mode standing rule so every Invoker plan the assistant writes sets executionAgent to Claude on each step, instead of leaving the field blank and inheriting Invoker’s default (the wrong agent).

The rule is documented next to the existing local-only plan flow bullet in SKILL.md, with matching corpus-lesson fixtures and a fourth unittest in test_artifact_sweeps.py that ties the skill text to those examples—same pattern as the local-only rule.

No Invoker runtime, hooks, or scripts change; behavior is instruction + test coverage only.

^{Reviewed by Cursor Bugbot for commit e9fcc23. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: e3e2f162-c628-4371-93fe-476e74c6677c)

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 637b9854-6918-477d-9337-64073fca486d)

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 433e38d7-84fc-4059-8c45-4b7b2553cef4)

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request