Create Gentle AI pull requests with issue-first checks. Trigger: creating, opening, or preparing PRs for review.
SkillScore92Maintained
Automatic
◆Coding pack
Last verified 10/7/2026
What it changes
Passed 22 of 24 test cases with the skill, and 13 of 24 without it.
Without skill54%
With skill92%
Checks it passed that the model alone missed
Assigns the type:breaking-change label
Specifies applying exactly one type:* label
Assigns the type:bug label for revert commits
and 7 more in the full report
9 fixed13 passed either way2 still failing0 broke
Pass-rate lift
+38 pp
Tokens
+111%
Agent turns
0%
Response time
-7%
Model gemini-3.6-flashMeasured 10/7/2026v522 of 24 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 9
Prompt
I formatted all bash scripts in the repository according to style guidelines. What exact PR type label maps to a `style` commit?
What was checked, without and with the skill
Checked
Without skill
With skill
Assigns the type:chore label for style commits
Missed
Passed
case-14 · gemini-3.6-flash · 10/7/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md165 linessingle file
When to Use
Use this skill when:
Creating a pull request for any change
Preparing a branch for submission
Helping a contributor open a PR
Critical Rules
Every PR MUST link an approved issue — no exceptions
Every PR MUST have exactly one type:* label
REQUIRED CI must pass according to target policy before merge; CodeRabbit is optional unless required by that policy
Blank PRs without issue linkage will be blocked by GitHub Actions
Target and Authorization
Inspect origin locally (git remote get-url origin) and establish one unambiguous target host and owner/repo. Do not infer the target from the cwd or assume main. Stop if ambiguous.
Obtain explicit remote destination, operation and credential/session authorization before any target-host reads, including gh auth status or repository metadata. Do not inspect or reuse an ambient SSH agent. Permission for a read does not authorize a write; confirm each remote operation is within the grant.
Once authorized, reuse fresh target-bound issue, default branch, type labels and checks evidence from this session instead of repeating discovery. Refresh stale or mismatched evidence; do not substitute another host's data. Resolve the base from the authorized target's default branch or a human-selected base.
Do not automatically commit, push, open a PR or merge. Prepare and report a candidate; perform each action only with its own authorization. Local preparation is not permission for remote delivery.
Workflow
Identify the authorized target and verify that the linked issue is approved; choose with the human whether the reference closes it or is nonclosing.
Select the base from target metadata and prepare a type/description branch only when requested.
Implement in work units with conventional commits when authorized; follow the merged ODD applicable test-first policy, run applicable tests, shellcheck on modified scripts, and test changed skills in at least one agent when relevant. The per-task advisory 400 authored-line heuristic is not an automatic split or a reason to omit tests or docs.
Prepare the PR body from the target template and evidence. On separate authorization, open the PR and add exactly one type:* label.
Check target-policy required CI and report pending/failing checks rather than declaring merge-ready.