eval-plan

garrytan/gstackAI & ML engineeringProduct & project management

eval-plan is a skill in garrytan/gstack, last changed 20 Nov 2026. It can touch credentials, run shell commands, pre-approve tools and run commands as it loads — read the findings before installing.

From the author
Preregister a paid eval (bars, arms, decoys, stop rules), dry-run it at zero cost, then run a three-item priced pilot whose spend table gates approval. (gstack)

At a glance

What it can touch

How much a skill can do, not whether it is risky.

Pattern check

3 items to review

Broad tools pre-approved in allowed-tools

R34 · High · SKILL.md:14and 2 more

Commit 0adfab9 · 20 Nov 2026

How the check works

Pattern check:3 items to reviewSee them

QAI Index

70/QI

We rate every listed skill the same way, from 0 to 100.

A guide to compare by, not a verdict on the skill.

How it's scored

How the QAI Index is scored

Adoption · up to 40
GitHub stars on the repo that holds the skill, shared among its skills, and copies of the skill in other repos.
Quality · up to 30
Our pattern check, the format check, a named licence and a description of a sentence or more.
Upkeep · up to 20
How recently the skill's folder changed.
Trust · up to 10
An official collection, or an organisation that has confirmed its domain with GitHub.

Get an email when its check changes.

Keep it with the other skills you use, installed with one command.

Install

Sign in to see how to install it

Free with Google or an email link, no password needed. Then pick your agent and copy its steps.

MIT
Licence
142.2k
Repo stars
Publisher
Files

What it can touch

Sign in to see the file and line behind each one

The summary at the top stays open to everyone. Signing in is free, with Google or an email link.

Pattern check

Pattern check: 3 items to review · commit 0adfab9 · 20 Nov 2026

We checked every text file in this skill, at commit 0adfab9 on 20 Nov 2026, against 45 published patterns such as download-and-run commands, hidden text and credential access. We don't run the skill, we don't read the web pages it points to, and we can't judge everything its instructions might lead an agent to do. In published tests, pattern checks like this one missed most malicious skills. Read the skill before you install it — especially anything it asks you to download or run.

How the check worksReport an incorrect finding

Format check

Each agent reads SKILL.md a little differently. We check the file against each one's published rules; we don't run the skill in any agent.

  • Agent Skills specPasses with warnings

    The format our Agent Skills docs set out, held to the strictest version of each rule among the agents we check.

    • Warning

      Fields outside the spec: preamble-tier, version and triggers. Uploads to the Claude apps and API, and strict validators, reject the file; most agents ignore these fields. O1

      Fix: Delete them for a portable skill, or keep them if you only target the agents that read them.

    • Warning

      allowed-tools is a list; the spec defines a space-separated string. gh skill publish rejects a list, though Claude Code accepts one. O5

      Fix: Write it on one line: allowed-tools: Bash Read Write Edit Glob Grep AskUserQuestion Bash(~/.claude/skills/gstack/bin/gstack-skill-load:*)

    • Warning

      The body is about 6,029 tokens (approx.); the spec suggests under 5,000. B2

      Fix: Move detail into reference files that the agent reads only when needed.

    • Note

      The description doesn't say when to use the skill. D6

      Fix: Add a sentence such as "Use when …" naming the requests or files that should trigger it.

    • Note

      There is no license field, so readers can't tell whether they may reuse the skill. O2

      Fix: Add license: with an SPDX name such as MIT or Apache-2.0, or the name of a bundled licence file.

    • Note

      Reference files over 100 lines with no table of contents at the top: sections/preamble.md (262 lines). B5

      Fix: Add a short list of the sections at the top of each.

    • Note

      Line 195 has dated wording ("before 2026"), which goes stale. B8

      Fix: Describe the current way, and move old patterns into their own clearly marked section.

  • Claude CodePasses

    Claude Code reads its own extra fields, ignores unknown ones, and still loads a skill whose frontmatter is broken, without its fields.

    • Note

      Claude Code ignores fields it doesn't know: preamble-tier, version and triggers. O1

      Fix: Check the spelling against Claude Code's field list, or delete them.

    • Note

      The body is about 6,029 tokens (approx.); the spec suggests under 5,000. After compaction Claude Code keeps only the first 5,000 tokens of a skill. B2

      Fix: Move detail into reference files that the agent reads only when needed.

    • Note

      Reference files over 100 lines with no table of contents at the top: sections/preamble.md (262 lines). B5

      Fix: Add a short list of the sections at the top of each.

    • Note

      Line 195 has dated wording ("before 2026"), which goes stale. B8

      Fix: Describe the current way, and move old patterns into their own clearly marked section.

  • Claude apps & API uploadFails

    Uploading to claude.ai or the Claude API, or packaging with Anthropic's script, accepts only the six spec fields and stops on anything else.

    • Error

      Upload and packaging fail on fields outside the spec: preamble-tier, version and triggers. O1

      Fix: Delete them, and keep your own data under metadata as text values.

    • Note

      allowed-tools is a list; the spec defines a space-separated string. O5

      Fix: Write it on one line: allowed-tools: Bash Read Write Edit Glob Grep AskUserQuestion Bash(~/.claude/skills/gstack/bin/gstack-skill-load:*)

    • Note

      The body is about 6,029 tokens (approx.); the spec suggests under 5,000. Anthropic's guidance says the same. B2

      Fix: Move detail into reference files that the agent reads only when needed.

    • Note

      Reference files over 100 lines with no table of contents at the top: sections/preamble.md (262 lines). B5

      Fix: Add a short list of the sections at the top of each.

    • Note

      Line 195 has dated wording ("before 2026"), which goes stale. B8

      Fix: Describe the current way, and move old patterns into their own clearly marked section.

  • OpenAI (Codex & API)Passes with warnings

    Codex finds SKILL.md by its exact name and needs a description; OpenAI's Skills API checks the frontmatter against the spec and caps uploads.

    • Warning

      Fields outside the spec: preamble-tier, version and triggers. Codex ignores them; OpenAI doesn't say whether its upload accepts them. O1

      Fix: Delete them, or move your own data under metadata as text values.

    • Note

      allowed-tools is a list; the spec defines a space-separated string. O5

      Fix: Write it on one line: allowed-tools: Bash Read Write Edit Glob Grep AskUserQuestion Bash(~/.claude/skills/gstack/bin/gstack-skill-load:*)

  • VS Code & CopilotPasses

    VS Code reads skills one folder below .github/skills, .claude/skills or .agents/skills and falls back to the folder name when the name is missing or differs; GitHub's Copilot docs require the exact name SKILL.md.

    • Note

      VS Code ignores fields it doesn't support: preamble-tier, version and triggers. O1

      Fix: Delete them unless another agent you target reads them.

    • Note

      allowed-tools is a list; the spec defines a space-separated string. O5

      Fix: Write it on one line: allowed-tools: Bash Read Write Edit Glob Grep AskUserQuestion Bash(~/.claude/skills/gstack/bin/gstack-skill-load:*)

SKILL.md

Files

Changelog

Newest first: each commit that changed its folder on GitHub, and each change to its listing here.

  1. Updated: Teach agents nine habits that several teams learned the hard way (GST… (ebc8674 (opens in a new tab))
  2. Updated: Telemetry records why skill runs end, which host and model ran them, … (9ea26ec (opens in a new tab))
  3. Updated: Claude Code skills run their start step as they load, so it can't be … (7f9c1bc (opens in a new tab))
  4. Updated: Ask each question once and check every handed-over file, in every ski… (a141e94 (opens in a new tab))
  5. Updated: Make gstack agents lead every report with what failed, what's unverif… (76fad3e (opens in a new tab))
  6. Updated: Load the never-print-secrets rule in every skill on every host, with … (28822af (opens in a new tab))
  7. Updated: Make /eval-plan check judges, zeros, failure counts and decisions bef… (fdf3f1e (opens in a new tab))
  8. Updated: Open findings and results write-ups with two or three plain takeaways… (a87d92c (opens in a new tab))
  9. Updated: v1.94.29.0 R56: everything in gstack earns its place (principle, evid… (ade35bc (opens in a new tab))
  10. Updated: v1.93.10.0 wave(docs): #3590 #3591 #3819 #3618 #3824 #3828 (#3847) (daf6613 (opens in a new tab))
  11. Updated: v1.93.6.0 feat: /plan-tune learns from ad-hoc question ids that carry… (5209e92 (opens in a new tab))
  12. Updated: v1.92.50.0 wave(autoplan): #3781 #3743 #3751 #3762 (#3788) (90a170c (opens in a new tab))
  13. Updated: v1.92.2.0 evals: no-gstack arms for the model panel, arm-benchmark, s… (4eff47c (opens in a new tab))
  14. Listed on QAI

QAI Index badge

Show this skill's QAI Index in a README or on a website. It links back to this page and changes when we refresh the catalog.

Preview of the QAI Index badge
[![QAI Index](https://qai.io/badge/skills/garrytan/gstack/eval-plan.svg)](https://qai.io/skills/garrytan/gstack/eval-plan)

More from garrytan/gstack

SkillQAI IndexWhat it can touchPattern checkLicenceUpdated
70/QI
  • Adds hooks
  • Touches credentials
  • +6 more
Pattern check: 3 items to reviewMIT19 Nov 2026
benchmarkWeb performance regression detection. Use for: web performance benchmark, check page speed. (gstack)
70/QI
  • Runs shell commands
  • +3 more

Pattern check: 2 items to reviewMITUpdated 19 Nov 2026

70/QI
  • Runs shell commands
  • Names web addresses
  • +2 more
Pattern check: 2 items to reviewMIT19 Nov 2026
benchmark-modelsCross-model benchmark for gstack skills. Use for: cross model benchmark, compare claude gpt gemini. (gstack)
70/QI
  • Touches credentials
  • +3 more

Pattern check: 2 items to reviewMITUpdated 19 Nov 2026

70/QI
  • Touches credentials
  • Runs shell commands
  • +2 more
Pattern check: 2 items to reviewMIT19 Nov 2026
canaryPost-deploy canary monitoring. Use for: monitor after deploy, canary check. (gstack)
70/QI
  • Runs shell commands
  • +3 more

Pattern check: 3 items to reviewMITUpdated 19 Nov 2026

70/QI
  • Runs shell commands
  • Reaches the network
  • +2 more
Pattern check: 3 items to reviewMIT19 Nov 2026
70/QI
  • Runs shell commands
  • Pre-approves tools
  • +1 more
Pattern check: 3 items to reviewMIT19 Nov 2026
codexOpenAI Codex CLI wrapper — three modes. Use for: codex review, second opinion, outside voice challenge. (gstack)
70/QI
  • Downloads software
  • +5 more

Pattern check: 3 items to reviewMITUpdated 19 Nov 2026

70/QI
  • Downloads software
  • Runs shell commands
  • +4 more
Pattern check: 3 items to reviewMIT19 Nov 2026
All skills in garrytan/gstack

Other AI & ML engineering skills

Same category, most repo stars first.

SkillQAI IndexWhat it can touchPattern checkLicenceUpdated
68/QI
  • Runs shell commands
Pattern check: Nothing flaggedMIT11 Aug 2026
68/QI
  • Runs shell commands
  • Names web addresses
Pattern check: Nothing flaggedMIT11 Aug 2026
claude-apiReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file).

anthropics/skills

72/QI
  • Touches credentials
  • +6 more

Pattern check: 1 item to reviewApache 2.0Updated 9 Oct 2026

72/QI
  • Touches credentials
  • Downloads software
  • +5 more
Pattern check: 1 item to reviewApache 2.09 Oct 2026
88/QI
  • Touches credentials
  • Runs shell commands
  • +3 more
Pattern check: Nothing flaggedApache 2.028 Oct 2026
70/QI
  • Runs shell commands
  • Reaches the network
  • +3 more
Pattern check: 1 item to reviewApache 2.020 Nov 2026
70/QI
  • Touches credentials
  • Runs shell commands
  • +4 more
Pattern check: 1 item to reviewApache 2.020 Nov 2026

Repo facts from GitHub on 20 Nov 2026. Files read at commit 0adfab9.

QAI did not write these skills and is not affiliated with their authors unless a page says so. Logos and icons belong to their owners and are not a mark of approval. How we check