CM-Stack is a set of Markdown skill files (skills/<name>/SKILL.md) plus two
context files (AGENTS.md, CLAUDE.md) that point agents at those skills.
There is no build step and no runtime dependency -- "installing" just means
getting those files somewhere your agent will read them, in a format it
understands.
Pick the section below that matches your harness. If your tool isn't listed, see Generic / unlisted harnesses.
git clone https://github.com/clearmeasure/cm-stack.git C:\tools\cm-stackAll instructions below assume you have a local copy of this repo. If you plan to customize skills for your team (per the README: "fork it, adapt the skills to your stack"), fork first and clone your fork instead.
Copilot CLI discovers skills from ~/.copilot/skills/<name>/.
git clone https://github.com/clearmeasure/cm-stack.git $env:USERPROFILE\.copilot\skills\cm-stackSkills appear as /clarity, /quality, /stability, /speed, /team,
/scorecard, /values-check in any session. No restart required for new
sessions; an already-running session won't pick up the change.
Live-editing note: if you're iterating on a skill file, clone once, then edit in place -- no symlink needed since Copilot CLI reads the file fresh each invocation.
Claude Code discovers skills from ~/.claude/skills/<name>/.
git clone https://github.com/clearmeasure/cm-stack.git $env:USERPROFILE\.claude\skills\cm-stackThis repo ships a .claude-plugin/plugin.json manifest, so it can also be
installed through Claude Code's plugin system if you prefer that flow (e.g.
for org-wide distribution via a plugin marketplace) rather than a bare skills
clone. Consult your Claude Code version's plugin-install docs for the exact
command, and point it at this repo's URL or local path.
Tools that don't have a discoverable skills directory but do support a
project- or user-level instructions/rules file (e.g. .cursorrules,
.windsurfrules, a custom system prompt) won't auto-discover skills/.
Wire them up manually:
- Clone CM-Stack somewhere stable (see Before you start).
- Copy the content of
AGENTS.md(orCLAUDE.md-- they're equivalent) into your tool's rules/instructions file, or@-reference/include it if your tool supports file includes. - Replace the relative links (
README.md,docs/FIVE_PILLARS.md,skills/<name>/SKILL.md) with the actual path to your CM-Stack clone. - When you want to invoke a skill, tell the agent explicitly, e.g.:
"Load
skills/quality/SKILL.mdfromC:\tools\cm-stackand follow it to review my last 5 merged PRs."
This works for any chat-based coding agent that can read files or take pasted instructions, even without native skill/plugin support.
If your agent supports any of these, use that mechanism:
- A skills or plugins directory -- clone/symlink this repo into it, following the pattern in the sections above.
- A project-level context file (
AGENTS.md,CLAUDE.md,.aiconfig, etc.) that's auto-loaded per session -- copy or reference this repo'sAGENTS.md/CLAUDE.md. - File attachments /
@-mentions in a prompt -- attach or reference the specificskills/<name>/SKILL.mdyou want to run for that session.
If none of those exist, fall back to manual invocation: paste the contents
of the relevant skills/<name>/SKILL.md directly into your prompt along with
your request. Every skill file is self-contained (frontmatter, instructions,
and example output), so this always works as a last resort.
If you're actively editing skills in your own local clone and want changes to show up immediately in an agent's skills directory without re-cloning:
# Windows (run as Administrator, or enable Developer Mode)
New-Item -ItemType SymbolicLink `
-Path $env:USERPROFILE\.copilot\skills\cm-stack `
-Target C:\tools\cm-stack\skills# macOS/Linux
ln -s /path/to/cm-stack/skills ~/.claude/skills/cm-stack- Cloned install:
git -C <clone-path> pull - Symlinked install: pull in your working copy; the symlink picks it up automatically.
- Copied-into-rules-file install: re-copy
AGENTS.md/CLAUDE.mdand any changedSKILL.mdcontent after pulling.
Check CHANGELOG.md after updating -- skill behavior/output changes are
recorded there per an entry's version bump in VERSION.
Regardless of harness, confirm the install worked before relying on it:
- Start a session inside a different project (not this repo).
- Ask the agent to run one skill, e.g. "Load CM-Stack, run
/clarityon this project" (or the manual-invocation phrasing for your harness). - Confirm the output roughly matches the "Example output" section in
skills/clarity/SKILL.md-- a scored table with findings, not a generic response. - If the agent doesn't recognize the skill or
/command, re-check the path it discovers skills/context from and confirm the clone/symlink/copy step above actually landed there (Get-ChildItemthe target directory).
| Symptom | Likely cause | Fix |
|---|---|---|
Agent doesn't recognize /clarity etc. |
Skill directory not on the agent's discovery path, or session started before install | Confirm the exact path your agent version scans (check its docs -- paths change between versions); start a fresh session after installing |
| Agent gives a generic answer instead of following the skill's structure | Skill was loaded as plain text, not as a skill/plugin | Explicitly tell the agent to "follow the instructions in SKILL.md exactly," or re-invoke using your harness's native skill/plugin mechanism |
| Symlink creation fails on Windows | No admin rights / Developer Mode disabled | Enable Developer Mode (Settings > For Developers) or run PowerShell as Administrator, or just use a plain clone instead |
Skills work but output is stale after a git pull |
Copied-into-rules-file install, or agent cached the file at session start | Re-copy the updated file, or restart the session |
Delete the clone/symlink from your agent's skills directory (or remove the copied section from your rules/context file). No other cleanup is required.