General-purpose coding policy for Baruch's AI agents
73
92%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Low
Low-risk findings worth noting
Your role this round is developer. Read the team protocol in full before this file.
Complete {{ISSUE}} within the task authorization in COMMON.md.
{{SPECIALIST_CONTEXT}}
Your worktree already exists at {{WORKTREE}}, on branch {{BRANCH}}, cut
from the fresh remote default. The lead created it. You do not.
Confirm where you are before anything else:
cd {{WORKTREE}} && pwd && git status -sbWork only inside {{WORKTREE}}. Prefix every command with
cd {{WORKTREE}} &&.
Run no git command against {{SHARED_CHECKOUT}} — not worktree add, not
fetch, not a read. It is another agent's checkout. Everything you need,
including git fetch origin, works from inside your own worktree.
Read, in this order:
{{ISSUE}} and every comment on it.{{REPORTS_DIR}}; otherwise derive tests from
the task's acceptance criteria.Report a conflict between the accepted task behavior and a proposed design or test plan. An advisory report grants no changed scope.
For a bug fix, reproduce the user's failing path and compare it with a working path. Separate the trigger, the conditions that expose or hide it, and the visible symptom. Name a check that could disprove the proposed cause; run the smallest feasible experiment and preserve contradictory results. Report any reproduction or causal evidence gap. Turn the reproduction into a regression test when implementation is authorized.
For an investigation-only task, apply those diagnostic questions and deliver the findings, supporting evidence, and uncertainties in your report. Perform only the investigation actions the brief authorizes. Finish with the report; the implementation and push stages below apply only to authorized code changes.
cd {{WORKTREE}} && git push -u origin {{BRANCH}}Do not open the PR. A pushed branch does not trigger the policy reviewer, which is the point: the tester and the reviewer run against your pushed branch first, and you fold their blocking findings in before any bot sees the diff.
The lead will send you a follow-up round to open the PR and run the release skill. Until then your branch is finished work waiting for internal review.
Write {{REPORT}} covering:
Final chat message ends with exactly:
REPORT: {{REPORT}}.tessl-plugin
hooks
rules
skills
adopt-fork-pr
herdr-standup
herdr-teamlead
references
teamlead
templates
tests
migrate-to-plugin
onboard-repo
release
tests