Health — collaboration stack (multi-platform)
Path matrix, host detection, commands → references/health-methods.md. scripts/collect-context.sh prints which platform marker files exist.
You may inline 🥷 at the start of the first reply.
Before running
- Open
health/references/health-methods.md: §2 identify host, then check the matching platform section from §5 onward.
- At repo root:
bash health/scripts/collect-context.sh "$(git rev-parse --show-toplevel)" — paste ## agent_markers and ## paths_present into report Facts; do not cat sensitive paths.
Multi-platform principles
- Same six-layer model (project docs → rules → skills/playbooks → hooks/automation → tools/MCP → secrets) for every host; paths differ.
- Coexistence:
.cursor/, CLAUDE.md, .windsurfrules, etc. may all exist — list what applies (ask which agent the user uses); flag duplication or conflict as [~].
- Unknown host: repo-generic checks +
agent_markers only; host-specific layers n/a, no penalty.
Teams using Claude Code with their own health collectors (e.g. collect-data.sh) may compare against this skill; this repo does not ship those scripts.
Six layers (quick → details in health-methods)
| Layer |
Examples |
| Project entry |
README.md, AGENTS.md, CLAUDE.md |
| Behavior rules |
.cursor/rules/, .claude/rules/, .windsurfrules, .clinerules, .github/copilot-instructions.md, .amazonq/rules/ |
| Skills / playbooks |
*/SKILL.md, Claude skills/, product project docs |
| Hooks / automation |
.cursor/hooks.json, IDE tasks (per product) |
| MCP / tools |
.cursor/mcp.json, Continue config, Claude settings.local.json |
| Secrets / local overrides |
.env, *.pem, settings.local — must not be committed |
Boundaries
- Stack configuration, not business logic correctness.
- No auto-edits;
Action: = copy-paste commands or explicit edit steps after user confirmation.
Output language
Follow project conventions (AGENTS.md, CLAUDE.md, Copilot instructions, platform rules); otherwise match the user's language.
Budget and depth
- Default summary:
collect-context.sh + sample key files + MCP list and harmless probes if possible.
- Deep audit: user asks thorough/complete, or
[!] found — expand first; warn about time and tokens.
Step 0: project tier (calibrate expectations)
| Tier |
Signals |
Expectations (relax for small repos) |
| Simple |
Small repo, no CI |
README or AGENTS; 0–1 rule files OK |
| Standard |
CI or small team |
README + repo rules; clear MCP/tool list |
| Complex |
Many packages, MCPs, hooks |
Layered rules, discoverable skills, least privilege, no secrets in git |
Do not grade Simple projects with Complex criteria.
Steps 1–4 (details in health-methods)
- Step 1 Facts: generic repo + per-host; prefer
collect-context.sh output.
- Step 2 MCP: harmless probes; skip
enabled: false; never paste full keys in the report.
- Step 3 Rules & skills: conflicts, duplication, over-broad globals,
name/description quality on */SKILL.md.
- Step 4 Report: template below.
Health Report: {project} ({Simple|Standard|Complex}, host: …)
### Facts (from collect-context or equivalent)
### [PASS] (table, ≤5 rows)
### Finding
- [severity] symptom (file:line if known)
Why: one line
Action: copy-paste command or concrete edit
Severity: [!] secrets / dangerous allowedTools / MCP abuse; [~] layering and duplication; [-] hygiene and docs.
If clean: All relevant checks passed. Nothing to fix.
Gotchas
| Situation |
Handling |
Missing python3/jq |
collect-context.sh notes json_skip; not a defect unless the project needs JSON CI |
| Hook "not firing" |
Check host debug / UI overlay before blaming config |
| External health scripts |
Use this repo's collect-context.sh as facts baseline; note conflicts side by side |
Source: zhijunio/zhijunio-skills — distributed by TomeVault.
1---2name: zhijunio-zhijunio-skills-health3description: Health — collaboration stack (multi-platform)4---56# Health — collaboration stack (multi-platform)78**Path matrix, host detection, commands** → **[`references/health-methods.md`](references/health-methods.md)**. **`scripts/collect-context.sh`** prints which platform marker files exist.910You may inline `🥷` at the start of the first reply.1112## Before running13141. Open **[`health/references/health-methods.md`](references/health-methods.md)**: **§2 identify host**, then check the matching platform section from **§5 onward**.152. At repo root: **`bash health/scripts/collect-context.sh "$(git rev-parse --show-toplevel)"`** — paste **`## agent_markers`** and **`## paths_present`** into report **Facts**; **do not** `cat` sensitive paths.1617## Multi-platform principles1819- **Same six-layer model** (project docs → rules → skills/playbooks → hooks/automation → tools/MCP → secrets) for every host; **paths differ**.20- **Coexistence**: `.cursor/`, `CLAUDE.md`, `.windsurfrules`, etc. may all exist — list **what applies** (ask which agent the user uses); flag **duplication or conflict** as `[~]`.21- **Unknown host**: repo-generic checks + `agent_markers` only; host-specific layers `n/a`, no penalty.2223Teams using **Claude Code** with their own health collectors (e.g. `collect-data.sh`) may compare against this skill; this repo does not ship those scripts.2425### Six layers (quick → details in health-methods)2627| Layer | Examples |28|-------|----------|29| Project entry | `README.md`, `AGENTS.md`, `CLAUDE.md` |30| Behavior rules | `.cursor/rules/`, `.claude/rules/`, `.windsurfrules`, `.clinerules`, `.github/copilot-instructions.md`, `.amazonq/rules/` |31| Skills / playbooks | `*/SKILL.md`, Claude `skills/`, product project docs |32| Hooks / automation | `.cursor/hooks.json`, IDE tasks (per product) |33| MCP / tools | `.cursor/mcp.json`, Continue config, Claude `settings.local.json` |34| Secrets / local overrides | `.env`, `*.pem`, `settings.local` — must not be committed |3536## Boundaries3738- **Stack configuration**, not business logic correctness.39- **No auto-edits**; `Action:` = copy-paste commands or explicit edit steps after user confirmation.4041## Output language4243Follow project conventions (`AGENTS.md`, `CLAUDE.md`, Copilot instructions, platform rules); otherwise match the user's language.4445## Budget and depth4647- **Default summary**: `collect-context.sh` + sample key files + MCP list and harmless probes if possible.48- **Deep audit**: user asks thorough/complete, or `[!]` found — expand first; warn about time and tokens.4950## Step 0: project tier (calibrate expectations)5152| Tier | Signals | Expectations (relax for small repos) |53|------|---------|--------------------------------------|54| **Simple** | Small repo, no CI | README or AGENTS; 0–1 rule files OK |55| **Standard** | CI or small team | README + repo rules; clear MCP/tool list |56| **Complex** | Many packages, MCPs, hooks | Layered rules, discoverable skills, least privilege, no secrets in git |5758**Do not grade Simple projects with Complex criteria.**5960## Steps 1–4 (details in health-methods)6162- **Step 1 Facts**: generic repo + per-host; prefer **`collect-context.sh`** output.63- **Step 2 MCP**: harmless probes; skip `enabled: false`; never paste full keys in the report.64- **Step 3 Rules & skills**: conflicts, duplication, over-broad globals, `name`/`description` quality on `*/SKILL.md`.65- **Step 4 Report**: template below.6667```text68Health Report: {project} ({Simple|Standard|Complex}, host: …)6970### Facts (from collect-context or equivalent)7172### [PASS] (table, ≤5 rows)7374### Finding75- [severity] symptom (file:line if known)76 Why: one line77 Action: copy-paste command or concrete edit78```7980**Severity**: `[!]` secrets / dangerous allowedTools / MCP abuse; `[~]` layering and duplication; `[-]` hygiene and docs.8182If clean: `All relevant checks passed. Nothing to fix.`8384## Gotchas8586| Situation | Handling |87|-----------|----------|88| Missing `python3`/`jq` | `collect-context.sh` notes `json_skip`; not a defect unless the project needs JSON CI |89| Hook "not firing" | Check host debug / UI overlay before blaming config |90| External health scripts | Use this repo's `collect-context.sh` as facts baseline; note conflicts side by side |9192---93> Source: [zhijunio/zhijunio-skills](https://github.com/zhijunio/zhijunio-skills) — distributed by [TomeVault](https://tomevault.io).94<!-- tomevault:4.0:skill_md:2026-05-23 -->
tomevault-io