Codex compatibility note:
- Invoke
$skill-name: load the registered .agents/skills/<name>/SKILL.md and execute it with Codex tools; no separate Skill tool or Claude mode switch is needed. When a workflow step or another skill names a skill, read that file in full yourself (every line, with no line limit; in parts only when output is cut off): a skill whose implicit invocation is off is never loaded for you. Author .claude/**; mirrors are generated.
- Track steps before execution and keep progress/evidence current. Explicit skill requests execute their protocol. Use the native question tool for required questions.
- Map foreign tool names to available host capabilities. A name mismatch needs no approval; an actually unavailable required capability stops with its step/evidence and a user question before adaptation.
- Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required
spawn_agent subagent(s) for that task.
- Preserve protocol order unless the user approves a deviation. Workflows use
$start-workflow: gates stay fixed; other steps may flex with logged reasons and evidence.
Quick Summary
Goal: Answer anything about (a) the portable .claude framework as configured for this project and (b) this project's own key technical, architecture, and structural knowledge — from generated output, never from memory.
Summary: Classify the question → run the matching generator → show its output and requested interpretation → hand off narrower help or requested changes. Keep project files unchanged and ground names, paths and counts in this session's command output.
Workflow:
- Classify the question against the routing table below.
- Run the matching generator command. Never answer from recall.
- Present the output verbatim, then add only the interpretation the user asked for.
- Route onward when the question belongs to a narrower help surface (
$project-config --help, $project-init --help).
Key Rules:
- MUST ATTENTION this skill is read-only and terminal. Run no scan, edit no project file, and never propose a change. Keep host task tracking for execution evidence. If the user wants a change afterwards, hand off to
$project-config or $project-init.
- MUST ATTENTION every number, path, and name in the answer comes from a command run in this session. The framework and the config both drift; a memorised count is a hallucination with a plausible shape.
- Show the generator output verbatim before commenting on it. Do not summarise it away.
- Config option questions ("what can I set", "who reads
docsRoots") belong to $project-config --help — delegate rather than duplicating.
- Never print secrets. The config holds references only; if a value looks like a credential, report the key and stop.
Be skeptical. Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence percentages (Idea should be more than 80%).
Scope
In scope: what the framework is, what each of its layers does, which skills/agents/workflows/hooks exist here, which reference docs exist and what each is for, how the mirrors are generated, this project's stack, structure, module map, and verification commands.
Out of scope: changing configuration ($project-config), initialising or re-initialising a project ($project-init), generic ClaudeKit command usage unrelated to this project ($framework-config --mode=help), and any implementation work.
Routing table
| The user asks… |
Run |
"what is this .claude thing / how is it wired / what are the layers" |
node .claude/skills/project-help/scripts/project-overview.cjs --framework |
| "what skills do I have / is there a skill for X" |
node .claude/skills/project-help/scripts/project-overview.cjs --skills (or --skills=<term>) |
| "what is the structure of this project / where does code live" |
node .claude/skills/project-help/scripts/project-overview.cjs --structure |
| "what stack / architecture / database / deployment" |
node .claude/skills/project-help/scripts/project-overview.cjs --stack |
| "how do I test / verify / what command do I run" |
node .claude/skills/project-help/scripts/project-overview.cjs --commands |
| "what docs exist / which doc do I read for X" |
node .claude/skills/project-help/scripts/project-overview.cjs --docs |
| "tell me everything" / no clear target |
node .claude/skills/project-help/scripts/project-overview.cjs --all |
| "what can I configure / who reads option X" |
node .claude/skills/project-config/scripts/project-config-help.cjs --overview — then delegate to $project-config --help |
| "every project-config option" / "what is inside option X, including nested and array-item fields" |
node .claude/skills/project-config/scripts/project-config-help.cjs --sections, then --section=<name> (or --search=<term>) |
"what can I set in .ck.json / .ck.local.json", "which environment variables", "how do I turn X off" |
node .claude/scripts/ck-config-help.cjs — then delegate to $framework-config --mode=help config |
| "which skills consume which option" |
node .claude/skills/project-config/scripts/project-config-help.cjs --consumers |
| "where do specs, plans, and ADRs live" (roots and tokens) |
node .claude/skills/project-config/scripts/project-config-help.cjs --roots |
| "what does init decide" |
$project-init --help |
A bare --skills=<term> argument filters the catalog; use it instead of guessing whether a skill exists.
The two generators
Both are plain node entrypoints using only node: built-ins (PORT-001). Neither takes a host package.json script, and both resolve the repository root by walking up for a marker, so they also run from the .agents/ mirror.
.claude/skills/project-help/scripts/project-overview.cjs — framework inventory and project knowledge. Reads .claude/.ck.json, the project-config file it points at, and the live contents of .claude/{skills,agents,workflows,hooks}. Modes: --framework, --skills[=<term>], --structure, --stack, --commands, --docs, --all (default).
.claude/skills/project-config/scripts/project-config-help.cjs — the configuration option surface. Reads the config SCHEMA, the portability tokens, and the reference-doc registry, then counts consumers by scanning the framework tree. Modes: --overview (default), --sections, --section=<name>, --consumers[=<key>], --docs, --roots, --skills, --current, --search=<term>, --json, --help.
Presentation rules
- Print the command you ran, then its output verbatim, then your interpretation — in that order, so the user can re-run it.
- Consumer counts are name-reference counts, not a call graph: a skill that mentions an option in prose counts. Say so whenever you quote one.
--current in the config helper reports declared-vs-defaulted only; validity is $project-config --validate, not this skill.
- When a doc is listed as configured but missing on disk, report the gap and name the owning
$scan --target=…; do not offer to create it here.
- If the user's real question is "change this", stop answering and route:
$project-config for an existing project, $project-init for an uninitialised one.
Closing Reminders
Goal: Answer framework and project-knowledge questions from current generated output.
MUST ATTENTION Route: classify → run generator → present output and requested interpretation → route narrower help or requested changes.
Keep the operation read-only and terminal. Track execution, verify names/counts/paths from the current command, redact secrets, and show the output before interpreting it.
1---2name: project-help3description: '[Utilities] Use when asking what the .claude framework does for THIS project: configuration, skills/agents/workflows, reference docs, key tech/architecture/structure facts. Command keyword search: framework-config --mode=help.'4disable-model-invocation: true5metadata:6 source: https://skillmd.com/skills/duc01226/project-help7---89> Codex compatibility note:10> - Invoke `$skill-name`: load the registered `.agents/skills/<name>/SKILL.md` and execute it with Codex tools; no separate `Skill` tool or Claude mode switch is needed. When a workflow step or another skill names a skill, read that file in full yourself (every line, with no line limit; in parts only when output is cut off): a skill whose implicit invocation is off is never loaded for you. Author `.claude/**`; mirrors are generated.11> - Track steps before execution and keep progress/evidence current. Explicit skill requests execute their protocol. Use the native question tool for required questions.12> - Map foreign tool names to available host capabilities. A name mismatch needs no approval; an actually unavailable required capability stops with its step/evidence and a user question before adaptation.13> - Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required `spawn_agent` subagent(s) for that task.14> - Preserve protocol order unless the user approves a deviation. Workflows use `$start-workflow`: gates stay fixed; other steps may flex with logged reasons and evidence.15## Quick Summary1617**Goal:** Answer anything about (a) the portable `.claude` framework **as configured for this project** and (b) this project's own key technical, architecture, and structural knowledge — from generated output, never from memory.1819**Summary:** Classify the question → run the matching generator → show its output and requested interpretation → hand off narrower help or requested changes. Keep project files unchanged and ground names, paths and counts in this session's command output.2021**Workflow:**22231. **Classify** the question against the routing table below.242. **Run** the matching generator command. Never answer from recall.253. **Present** the output verbatim, then add only the interpretation the user asked for.264. **Route onward** when the question belongs to a narrower help surface (`$project-config --help`, `$project-init --help`).2728**Key Rules:**2930- MUST ATTENTION this skill is **read-only and terminal**. Run no scan, edit no project file, and never propose a change. Keep host task tracking for execution evidence. If the user wants a change afterwards, hand off to `$project-config` or `$project-init`.31- MUST ATTENTION every number, path, and name in the answer comes from a command run **in this session**. The framework and the config both drift; a memorised count is a hallucination with a plausible shape.32- Show the generator output verbatim before commenting on it. Do not summarise it away.33- Config **option** questions ("what can I set", "who reads `docsRoots`") belong to `$project-config --help` — delegate rather than duplicating.34- Never print secrets. The config holds references only; if a value looks like a credential, report the key and stop.3536**Be skeptical. Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence percentages (Idea should be more than 80%).**3738## Scope3940**In scope:** what the framework is, what each of its layers does, which skills/agents/workflows/hooks exist here, which reference docs exist and what each is for, how the mirrors are generated, this project's stack, structure, module map, and verification commands.4142**Out of scope:** changing configuration (`$project-config`), initialising or re-initialising a project (`$project-init`), generic ClaudeKit command usage unrelated to this project (`$framework-config --mode=help`), and any implementation work.4344## Routing table4546| The user asks… | Run |47|---|---|48| "what is this `.claude` thing / how is it wired / what are the layers" | `node .claude/skills/project-help/scripts/project-overview.cjs --framework` |49| "what skills do I have / is there a skill for X" | `node .claude/skills/project-help/scripts/project-overview.cjs --skills` (or `--skills=<term>`) |50| "what is the structure of this project / where does code live" | `node .claude/skills/project-help/scripts/project-overview.cjs --structure` |51| "what stack / architecture / database / deployment" | `node .claude/skills/project-help/scripts/project-overview.cjs --stack` |52| "how do I test / verify / what command do I run" | `node .claude/skills/project-help/scripts/project-overview.cjs --commands` |53| "what docs exist / which doc do I read for X" | `node .claude/skills/project-help/scripts/project-overview.cjs --docs` |54| "tell me everything" / no clear target | `node .claude/skills/project-help/scripts/project-overview.cjs --all` |55| "what can I configure / who reads option X" | `node .claude/skills/project-config/scripts/project-config-help.cjs --overview` — then delegate to `$project-config --help` |56| "every project-config option" / "what is inside option X, including nested and array-item fields" | `node .claude/skills/project-config/scripts/project-config-help.cjs --sections`, then `--section=<name>` (or `--search=<term>`) |57| "what can I set in `.ck.json` / `.ck.local.json`", "which environment variables", "how do I turn X off" | `node .claude/scripts/ck-config-help.cjs` — then delegate to `$framework-config --mode=help config` |58| "which skills consume which option" | `node .claude/skills/project-config/scripts/project-config-help.cjs --consumers` |59| "where do specs, plans, and ADRs live" (roots and tokens) | `node .claude/skills/project-config/scripts/project-config-help.cjs --roots` |60| "what does init decide" | `$project-init --help` |6162A bare `--skills=<term>` argument filters the catalog; use it instead of guessing whether a skill exists.6364## The two generators6566Both are plain `node` entrypoints using only `node:` built-ins (PORT-001). Neither takes a host `package.json` script, and both resolve the repository root by walking up for a marker, so they also run from the `.agents/` mirror.6768- **`.claude/skills/project-help/scripts/project-overview.cjs`** — framework inventory and project knowledge. Reads `.claude/.ck.json`, the project-config file it points at, and the live contents of `.claude/{skills,agents,workflows,hooks}`. Modes: `--framework`, `--skills[=<term>]`, `--structure`, `--stack`, `--commands`, `--docs`, `--all` (default).69- **`.claude/skills/project-config/scripts/project-config-help.cjs`** — the configuration option surface. Reads the config SCHEMA, the portability tokens, and the reference-doc registry, then counts consumers by scanning the framework tree. Modes: `--overview` (default), `--sections`, `--section=<name>`, `--consumers[=<key>]`, `--docs`, `--roots`, `--skills`, `--current`, `--search=<term>`, `--json`, `--help`.7071## Presentation rules7273- Print the command you ran, then its output verbatim, then your interpretation — in that order, so the user can re-run it.74- Consumer counts are **name-reference counts**, not a call graph: a skill that mentions an option in prose counts. Say so whenever you quote one.75- `--current` in the config helper reports declared-vs-defaulted only; validity is `$project-config --validate`, not this skill.76- When a doc is listed as configured but missing on disk, report the gap and name the owning `$scan --target=…`; do not offer to create it here.77- If the user's real question is "change this", stop answering and route: `$project-config` for an existing project, `$project-init` for an uninitialised one.7879---8081## Closing Reminders8283**Goal:** Answer framework and project-knowledge questions from current generated output.84**MUST ATTENTION Route:** classify → run generator → present output and requested interpretation → route narrower help or requested changes.85Keep the operation read-only and terminal. Track execution, verify names/counts/paths from the current command, redact secrets, and show the output before interpreting it.