minion-orchestrator

Unified Minions skill for both deterministic shell jobs and LLM subagent orchestration. Replaces the older `gbrain-jobs` routing intent. Use when: submitting gbrain jobs, shell/background tasks, spawning subagents, checking progress, steering running work, pausing/resuming, parallel fan-out. One durable, observable, steerable queue interface. Also carries the durable-execution doctrine for any operation expected to exceed ~2 minutes: capability ladder, deadman checks that verify the result was reported, and content-addressed stage checkpoints for expensive pipelines.

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

85%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.

A highly actionable, well-sequenced body with executable commands, explicit validation checkpoints, and excellent coverage of both job lanes and the durable-execution ladder. Its main weaknesses are a monolithic single-file structure with no progressive disclosure into reference files (and a project docs dependency instead), plus deliberate but costly repetition of the deadman doctrine across three sections.

Suggestions

Split the durable-execution material into a reference file (e.g. references/durable-execution.md) — the capability ladder, deadman pattern, and stage-checkpoint appendix — and keep SKILL.md to the routing table plus a one-paragraph summary with clear links, reducing the ~480-line body substantially.

Move the brain-tools allowlist enumeration and the full flag inventories (gbrain jobs submit / gbrain agent run) into a short reference file, keeping only the 2-3 most common tools and flags inline.

Consolidate the deadman-doctrine statements: state the rules once (Contract or Durable execution) and have the other two sections reference them by link/anchor instead of restating the same three failure modes.

Dimension	Reasoning	Score
Conciseness	The body is dense and nearly every line carries project-specific facts Claude cannot know (the GBRAIN_ALLOW_SHELL_JOBS gate, MCP trust boundary, coalescing behavior, lock TTL semantics, flag inventories) — there is no padding explaining generic concepts. It misses 5 because the durable-execution doctrine is stated three times (Contract, Durable execution section, Anti-Patterns) — an acknowledged deliberate mirror, but still duplication that could be consolidated.	4 / 5
Actionability	Guidance is copy-paste executable throughout: full CLI invocations ('gbrain jobs submit shell --params "{\"cmd\":\"echo hello\",\"cwd\":\"/abs/path\"}"'), argv forms, monitor/control command blocks, the fanout-manifest invocation, lifecycle ops with parameter examples (replay_job with data_overrides), and a concrete timeout table for common long operations. Common cases (submit, monitor, steer, cancel/replay, long-op routing) are each covered by a specific example.	5 / 5
Workflow Clarity	Multi-step processes are explicitly sequenced — Phases 1-5 for subagent jobs, the three-rung capability ladder ordered by deployment capability, and a numbered deadman pattern — with explicit validation checkpoints throughout: 'run gbrain jobs stats to confirm the worker is registered', the deadman's reported-check with a four-branch decision tree (reported/unreported/still-running/dead), checkpoint freshness checks, integrity-on-restore checks, and disarm-on-completion feedback. This is batch/destructive-adjacent work and validation is present, so no cap applies.	5 / 5
Progressive Disclosure	No bundle files exist (no references/, scripts/, or assets/), so all ~480 lines live in SKILL.md itself. Section headers and a routing table give reasonable navigation, but content that clearly belongs in separate one-level-deep reference files is inlined — the content-addressed stage-checkpoint appendix, the 14-tool brain allowlist enumeration, and the full flag inventory — matching the level-3 anchor ('some structure... content that should be separate is inline') rather than 4, where most such content would be split out.	3 / 5
	Total	17 / 20 Passed

Description

88%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.

A strong description: third-person, explicitly split into what it does and a 'Use when:' trigger list, with comprehensive coverage of both the shell-job and subagent lanes plus the durable-execution doctrine. Minor risks are jargon-heavy phrasing in the doctrine summary and trigger overlap with generic background-task skills.

Dimension	Reasoning	Score
Specificity	The description lists multiple concrete, distinct actions — 'submitting gbrain jobs, shell/background tasks, spawning subagents, checking progress, steering running work, pausing/resuming, parallel fan-out' — plus concrete mechanisms ('deadman checks that verify the result was reported', 'content-addressed stage checkpoints'), giving comprehensive coverage of the skill's two lanes. It is not a level-4 case because no meaningful capability lane is left unenumerated; the only near-gap (durable-execution detail) is explicitly named rather than omitted.	5 / 5
Completeness	It explicitly answers 'what' ('Unified Minions skill for both deterministic shell jobs and LLM subagent orchestration... One durable, observable, steerable queue interface') and 'when' with an explicit 'Use when:' clause enumerating concrete trigger cases (submitting, spawning, checking progress, steering, pausing/resuming, fan-out, >2-minute operations). This matches the level-5 anchor verbatim in structure — both what and when, with concrete trigger phrases.	5 / 5
Trigger Term Quality	The description embeds natural phrases users would say ('run in background' is echoed by 'shell/background tasks', 'spawn agent' by 'spawning subagents', 'pause agent' by 'pausing/resuming', 'parallel tasks' by 'parallel fan-out'), and the frontmatter also carries a dedicated 26-entry triggers list ('run in background', 'what's running', 'the job went silent', 'do these in parallel'). It falls short of 5 because the description itself leans on project jargon ('durable-execution doctrine', 'capability ladder', 'steerable queue interface') that a user would not naturally say verbatim.	4 / 5
Distinctiveness Conflict Risk	The skill occupies a clear niche (gbrain/Minions job queue and durable execution) with product-specific triggers ('submitting gbrain jobs', 'gbrain agent run') that no other skill would claim. It is not a 5 because the description's broader phrases ('shell/background tasks', 'spawning subagents', 'parallel fan-out') overlap with generic background-execution and subagent-orchestration skills and could fire on requests that don't involve this queue.	4 / 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
skill_md_line_count	SKILL.md is long (533 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: garrytan/gbrain
Path: skills/minion-orchestrator/SKILL.md
Commit: e78f1c3

Reviewed: 5 days 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.