zby/commonplace/cp-skill-health-check
Claude Code and agent skill by zby/commonplace
- community
- ext:[{"type":"config","description":"Configuration overrides"}]
Checking that the SKILL.md file is still in the repository.
Diagnose Commonplace installation, skill discovery, command PATH, and workflows blocked by unavailable sub-agents, nesting depth, or context isolation.
Install
npx vskill@latest install zby/commonplace/cp-skill-health-checkSemi-extensibleMentions customization, but needs skill-specific knowledge. Read the extensible skills standard
SKILL.md
Commonplace Health Check
EXECUTE NOW
Identify the installation or runtime constraint blocking a Commonplace operation and report the smallest repair supported by evidence.
Use this skill when a Commonplace KB has missing skills, missing or failing commonplace-* commands, commands that work in one shell but not in an IDE or agent runtime, or a workflow blocked on sub-agent creation or isolation. It applies both to installed projects, which read the Commonplace library from the installed package through pointers commonplace-init writes, and to the Commonplace source repository, whose own kb/ is the library.
Target symptom: $ARGUMENTS
Do not modify files during diagnosis. Report the evidence and the smallest repair. Apply a repair only when the user asked for one.
Run in the context being diagnosed; do not delegate the health check itself. If already invoked in a worker, identify that scope and do not extrapolate its capabilities to its parent. For a delegation symptom, run Step 3 after the layout and skill checks; run the package checks only if the evidence also points to an installation or command failure. For a general health check, run all steps.
Step 1 — Locate the project and classify the layout
From the workspace root, inspect:
Interpretation:
- Installed project:
.commonplace/library.mdexists and names the library root in the installed package. The project holds no library files. - Source repository:
pyproject.toml,src/commonplace/, and top-levelkb/{theory,reference,instructions}/exist. Itskb/is the library. - Old library copy:
kb/commonplace/comes from a release that copied the library into projects. Report it as requiring separate cleanup.commonplace-initleaves these files and their review baselines untouched; inspect local changes and references before removing copies. - Neither: the wrong directory is open or
commonplace-inithas not run. .venvis not a Commonplace layout requirement. A project may still own one for unrelated dependencies.
Step 2 — Check the control plane, library pointers, and skills
Read the active root control-plane file and confirm it routes Commonplace commands by bare name and, in an installed project, points agents to .commonplace/library.md. Then check init's pointers into the library and the runtime skills:
Interpretation, for an installed project:
- Every line
ok: the skill stubs,.commonplace/library.md, and the Claude Code read rule match the installed library. staleormissing: the library changed or moved since init last ran, for example after an upgrade that changed the skill set or a switch between editable and normal installs. Reruncommonplace-init. Until then, agents may read an outdated copy of the library without any warning.foreign: a skill directory thatcommonplace-initdid not write shadows the library's skill. Remove it after checking it holds nothing the user needs, then reruncommonplace-init.collision: a file in the project'skb/types/has the same path as a library global type, so every artifact typedtypes/<name>.mdfails validation. It is usually a leftover copy from an older release. Delete it, or move its change into a collection's owntypes/directory.- Reads of library files denied in Claude Code: the read rule in
.claude/settings.local.jsonis missing or names another library root;commonplace-init --checkreports it.
In the source repository, .claude/skills/ and .agents/skills/ hold committed symlinks into kb/instructions/; commonplace-init --check does not apply. Do not run commonplace-init there; follow the source repository's AGENTS.md instead.
Step 3 — Check delegation, agent depth, and isolation
Inspect the current session's exposed tools, runtime instructions, effective settings when available, and any supplied launch error. Record:
- whether a harness worker-launch tool is available and permitted;
- the current agent depth, the maximum permitted depth, and their counting
convention, when exposed;
- how many additional nested worker levels those values permit;
- whether launch supports a fresh context without the parent's conversation;
- available concurrent worker capacity, separately from nesting depth.
Report unexposed values as unknown. A configuration file is evidence of configured intent, not proof that the running session loaded it. Neither an installed agent CLI nor a skill's allowed-tools declaration proves that a worker-launch tool is available. Do not infer depth from agent names, the number of visible chats, or the number of free worker slots.
For a failing workflow, read its operative instructions and map the nesting needed from the failing caller: who launches whom, and at which stage. Count a skill's context: fork only if this runtime actually executes it in a child context. Sequential sibling workers reuse a level; do not add their count to the required depth. For ingest, the agent executing cp-skill-ingest must be able to launch a fresh drafting worker; if ingest itself runs in a child of the user-facing agent, drafting needs a further level. Compare the required nesting with the remaining depth at that caller, not merely at the health-check agent. If the caller's depth or runtime behavior is unavailable, report that comparison as unresolved.
When a delegation symptom remains unresolved and a harness launch tool is available and permitted, make at most one diagnostic worker launch from the current context. Use a fresh-context option if supported. Give it only:
Collect the response and release the worker through the harness when supported. Preserve any launch failure verbatim. Success establishes only that one child could run from this context at this time; it does not establish the maximum depth or prove conversation isolation. Establish isolation from the launch API's documented contract and the actual launch arguments; otherwise report it as unknown. Do not probe recursively or launch an agent CLI from the shell. If probing is unavailable or prohibited, report the evidence gap and continue diagnosis.
Classify the evidence as missing launch capability, delegation prohibited, depth exhausted, concurrent capacity exhausted, isolation unavailable, another launch failure, or unresolved. A missing tool alone does not establish why it is missing. Recommend a shallower entry point, enabling supported delegation, or waiting for an owned worker slot only when the evidence supports that repair. Report the specific workflow stage affected. A diagnosis does not authorize changing runtime settings or bypassing the workflow's isolation rule.
Step 4 — Check the uv tool installation and command ownership
On Linux/macOS:
On native Windows PowerShell:
If the bare name resolves to a .cmd file, or in Git Bash to an extensionless script, read that one file: Get-Content (Get-Command commonplace-validate).Source in PowerShell, or cat "$(command -v commonplace-validate)" in Git Bash.
Classify the result:
- uv missing: the installation prerequisite is absent.
- `llm-commonplace` absent from `uv tool list`: the user-level tool is not installed.
- Command exists inside `uv tool dir --bin` but bare-name lookup fails: the tool executable directory is missing from this process's
PATH. - Bare name resolves to a supported Windows shim: the file is a single line that runs
<uv tool dir>\llm-commonplace\Scripts\python.exe -con the same command's entry point, as written by the shim snippet inINSTALL.md(/c/...path form in the Git Bash script). Count it as the uv tool, not a conflict: Windows installs require these shims ahead ofuv tool dir --bin. If thatpython.exeis missing, the tool environment is damaged; reinstall it and rerun the snippet.commonplace-initmakes the same distinction. - Bare name resolves outside `uv tool dir --bin` otherwise: another install, a project venv, or a shim for another interpreter or command shadows the user-level tool. Report every candidate from
which -aorGet-Command -All. - Only some declared `commonplace-*` commands exist: the tool installation is incomplete or stale after an entry-point metadata change; reinstall it.
- A fresh terminal resolves the command but the IDE or agent does not: the launch class did not inherit the updated user environment. Treat this as runtime environment propagation, not package absence.
Do not use uv tool install --force as the default repair for an executable conflict. First identify which installation owns the conflicting name and remove or reorder the stale owner.
Step 5 — Check legacy project-environment residue
Inspect only; do not delete. Do not print or read .envrc into the conversation: it may hold secrets. Compare it with the file older Commonplace releases generated and report only the classification.
The exact .envrc generated by older Commonplace releases contained only:
POSIX shell:
On native Windows PowerShell:
The exact legacy file is obsolete after the user-level tool works in fresh processes. A differing .envrc needs manual review by the operator; tell them it exists and differs, and leave the reading to them. A .venv may hold the project's own dependencies; never classify it as removable merely because Commonplace no longer needs it.
Step 6 — Check package and validator health
If this is an installed project:
If this is the Commonplace source repository:
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingkb/instructions/cp-skill-health-check/SKILL.md
Popularity
- Repository stars
- 82
- Repository forks
- 10
- 7-day trend score
- 22
Stars and forks are for the whole GitHub repository, not this skill alone. Refreshed 6 Aug 2026, 3 months ago.
Extension points
configConfiguration overrides
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/zby/commonplace/cp-skill-health-check)