SuperAPI AI Directory

project-scan

Agents & models

View repository

Updated 10-15 21:33

About this skill

Project Scan

When the user explicitly requests project-scan, scanning the project, generating project-level AGENTS.mdgenerating or refreshing README.md, or refreshing the project-level Codex context, use this skill.

Do not use for:

  • feature development or bug fixes
  • overwriting existing AGENTS.md、README.md、Dockerfile or scripts without confirmation

List of files that must be generated

After scanning is complete, by default fill in the project scaffolding all at once according to the following list (for each existing one, ask "overwrite / skip / incremental update"):

No. File Purpose
1 AGENTS.md Project-level Codex context
2 README.md Project description
3 restart.sh Frontend and backend packaging + startup script
4 .claudeignore Claude Code ignore
5 .geminiignore Gemini CLI ignore (content copied from .claudeignore)
6 .gitignore Git ignore
7 .dockerignore Docker ignore
8 Dockerfile Container build
9 docker-compose.yml Container orchestration

The list cannot be reduced; align with the output of the same-named command in Claude /project-scan, Cursor.

Core approach

  1. First scan the key signal files and directories in the repository, distinguishing facts that are "clearly identified" from "inferences based on structure".
  2. By default ensure project-level .codex/tasks/、.codex/tasks/archived/、.codex/templates/ exists; if writing to the project root triggers a permission boundary, go through platform approval first, do not silently skip.
  3. Read the existing root directory AGENTS.md、README.md、.claude/CLAUDE.md, and reuse cc-project-init/assets/AGENTS.project.md as the AGENTS.md template source.
  4. First output a scan summary (including a tech stack identification table + file existence table), then ask 3-5 key questions: one-sentence project description, whether to generate the full scaffolding set, handling strategy for existing files (overwrite / incremental / skip), Docker exposed ports, API contract confirmation.
  5. AGENTS.md / README.md writing strategy:
    • Does not exist → generate based on template
    • Already exists and contains <!-- AUTO:* --> block → prioritize incremental update of the automatic block, preserve handwritten content
    • Already exists but has no automatic block → must first let the user choose "first-time full rebuild / append automatic block / skip"
  6. Scaffolding file writing strategy(restart.sh / ignore / Dockerfile / docker-compose):
    • Does not exist → from cc-project-init/assets/ select a template, render it according to the tech stack, and write it to disk
    • Already exists → ask "overwrite / skip", do not silently overwrite
    • .geminiignore content must be consistent with .claudeignore Stay consistent
  7. restart.sh after writing to disk must chmod +x。
  8. For template selection rules see cc-project-init/references/template-selection.md; for ignore file content specifications see references/scaffold-files.md。
  9. After completion output a summary: generated files, skipped files, completed .codex skeleton, and project facts that still need to be supplemented by the user.

Collaboration constraints

  • Project-level AGENTS.md template asset reuse cc-project-init/assets/AGENTS.project.md
  • Scaffolding template reuse cc-project-init/assets/(restart-go.sh.tmpl、Dockerfile-go.tmpl、docker-compose-*.yml.tmpl、gitignore-go.tmpl、dockerignore.tmpl etc.)
  • README.md generation and updates follow references/readme-update.md
  • Subsequent feature task persistence depends on new-feature
  • If it involves interface contract summarization or judgment of list/pagination/filter structures, follow cc-api-contract-safety in a way that does not speculate about the response structure

Output requirements

  • Clearly distinguish between "scan recognition" and "items to be confirmed"
  • Clearly state which of "create new / incremental update / first-time full rebuild / append automatic block / overwrite / skip" is used for each file
  • If completed .codex/tasks/、.codex/tasks/archived/、.codex/templates/, it must be clearly stated
  • If the user chooses to skip an existing file, explain the subsequent risks
  • If the interface contract cannot be determined from code or documentation, in AGENTS.md mark "pending user confirmation"; do not fabricate specifications
  • If the startup method, port, deployment method, or license cannot be reliably confirmed,README.md must be marked "pending confirmation"; do not fabricate commands
  • If the repository has no Cargo.toml、go.mod、pom.xml、package.json one of which can infer the tech stack, you must first confirm the target tech stack with the user, then choose a template, to avoid generating scaffolding for the wrong stack

Expand as needed

  • Scan items:references/scan-checks.md
  • Automatic block:references/agents-blocks.md
  • Rebuild strategy:references/rebuild-strategy.md
  • README update:references/readme-update.md
  • Scaffold file list and ignore file specifications:references/scaffold-files.md
  • Template selection:cc-project-init/references/template-selection.md