Name: shepherd-driver
Rating: 67.2 (1 reviews)
Author: microsoft

shepherd-driver

Use only as the composed drive-to-merge stage of an APM batch orchestrator (batch-bug-shepherd, apm-issue-autopilot) that has already selected ONE open pull request in microsoft/apm. Do NOT use for user-facing requests to triage issues, sweep a queue, or open PRs -- the parent orchestrator owns those. Spawn one shepherd-driver subagent per PR: it classifies copilot-pull-request-reviewer[bot] inline review, runs the apm-review-panel, folds (by default) every recommendation inside the PR's stated scope, pushes to the head branch or a superseding PR that preserves authorship via commit trailers, watches CI to green, and iterates under fixed caps until ready-to-merge, advisory-with-deferred, superseded, or blocked. Also provides the cross-PR conflict-resolution and mergeability-gate phase. This is NOT a standalone entrypoint.

Quality

82%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

77%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is a well-structured orchestration overview with an explicitly sequenced, validation-rich convergence loop and exemplary progressive disclosure via real bundle files. The main weakness is redundancy: the owner-touch gate and fold-by-default rules are described in multiple sections and could be consolidated for token efficiency.

Suggestions

Consolidate the canonical-owner gate and fold-by-default descriptions into a single authoritative section; have the convergence-loop phases reference it instead of restating it.

Trim the genesis-packet framing ('genesis R3 EXTRACT', 'A9 SUPERVISED EXECUTION', 'truth #2 CONTEXT EXPLICIT') where it does not change what the agent must do.

Dimension	Reasoning	Score
Conciseness	The body is dense and mostly load-bearing rather than explaining concepts Claude already knows, but the canonical-owner gate and fold-by-default invariants are restated across 'Convergence loop contract', 'Disciplines enforced', and the boundary section, so it could be tightened by consolidating.	3 / 5
Actionability	Provides concrete executable artifacts (the dependency-probe shell command, the exact ruff lint pair, the owner_touch_gate.py CLI with --help) and points to specific spawn-body/schema assets, with only minor gaps because the bulk of execution lives in the referenced bundle files.	4 / 5
Workflow Clarity	The convergence loop is an explicitly numbered X.0-X.8 sequence with validation checkpoints (FAIL-CLOSED owner gate, mutation-break gate, lint-until-silent, CI watch+recovery), explicit hard caps, and feedback loops/checklists for a destructive/batch operation.	5 / 5
Progressive Disclosure	SKILL.md is a clear overview with one-level-deep, well-signaled references (assets/, scripts/, references/mergeability-gate.md) all verified to exist as real files, a dedicated 'Bundled assets' index, and load-on-demand guidance for the conflict-resolution phase.	5 / 5
	Total	17 / 20 Passed

Description

87%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is highly specific and complete, clearly delimiting both what the skill does and when (and when not) to use it, with strong distinctiveness. Its only weakness is the imperative/second-person voice ('Use only as...'), which the rubric penalizes on specificity.

Suggestions

Rewrite the opening in third person (e.g. 'Drives a single selected PR to a merge-ready state as the composed stage of an APM batch orchestrator') to avoid the second-person voice penalty.

Add a couple of natural-language trigger phrases a user/orchestrator might actually say so the trigger-term quality reaches full coverage.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions (classify Copilot review, run apm-review-panel, fold recommendations, push to head/superseding PR, watch CI to green, iterate under caps) which is comprehensive coverage; reduced from 5 because the imperative 'Use only as...' is second-person voice, which the rubric penalizes by one point.	4 / 5
Completeness	Explicitly answers both 'what' (composed drive-to-merge stage performing the listed actions plus conflict-resolution/mergeability-gate) and 'when' ('Use only as the composed drive-to-merge stage of an APM batch orchestrator that has already selected ONE open PR' plus a 'Do NOT use for...' anti-trigger).	5 / 5
Trigger Term Quality	Good keyword coverage for its niche (drive-to-merge, APM batch orchestrator, batch-bug-shepherd, apm-issue-autopilot, pull request, CI, mergeability), but the terms are internal/orchestrator jargon rather than the natural phrases an end user would say.	4 / 5
Distinctiveness Conflict Risk	Clear niche (per-PR convergence loop for named orchestrators on microsoft/apm) with distinct triggers, explicit negative boundaries, and a 'NOT a standalone entrypoint' statement, giving minimal conflict risk.	5 / 5
	Total	18 / 20 Passed

Validation

87%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
relative_links	Relative link issues: 2 suspicious	Warning
referenced_paths_exist	Referenced path issues: 1 missing	Warning

	Total	14 / 16 Passed

Repository: microsoft/apm
Path: .agents/skills/shepherd-driver/SKILL.md
Commit: 71a877c

Reviewed: ~4 hours ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.