Context7

Context7 Docs + Skill Wizard

Retrieve current external library documentation via Context7 so implementation guidance is grounded in current docs instead of memory, and support explicit Context7 CLI requests (ctx7 library|docs|skills|setup|login|whoami).

When to use

• Use this skill when the user needs current external library or framework documentation.
• Use this skill when the user explicitly asks for Context7 CLI flows such as ctx7 skills generate, ctx7 skills install, ctx7 skills suggest, or ctx7 setup.
• Use this for version-sensitive API behavior, dependency troubleshooting, and skill-wizard install target guidance.
• Use this when docs lookup should run CLI-first with MCP as a backup retrieval path.
• Do not use it for OpenAI platform docs; route those to openai-docs.

Philosophy

• Prefer current-source documentation over memory for drift-prone dependency behavior.
• Run CLI-first with explicit command safety and fallback behavior.
• Keep answers implementation-shaped and directly tied to the user’s concrete task.

Required inputs

• For docs lookup: library or product name, implementation question, optional version constraints.
• For skill wizard: requested action (search|install|list|remove|suggest|info|generate|setup|login|whoami|logout) and target scope.
• For secure CLI execution: prefer op run --env-file ~/.codex/.env -- ctx7 ....
• For exact flag mapping and command forms, use references/context7-skill-wizard.md.

Deliverables

• Docs lookup outputs:

1. Resolved Context7 library id.
2. Focused docs-backed answer tied to the user question.
3. Source basis (documentation source or retrieval context).
4. Explicit assumptions or ambiguity notes when needed.
5. Execution path used (cli_primary, mcp_backup, or api_backup) and fallback reason when not CLI-primary.

• Skill wizard outputs:

1. Exact ctx7 command(s).
2. Selected target flags and scope (project vs global).
3. Source basis (command documentation or CLI reference).
4. Post-install verification commands and restart reminder when applicable.

Workflow

Lane A: Docs lookup (CLI primary, MCP/API backup)

1. Run CLI-first using secure env injection:
   • op run --env-file ~/.codex/.env -- ctx7 library <name> "<query>" --json
   • op run --env-file ~/.codex/.env -- ctx7 docs <libraryId> "<query>" --json
2. If CLI is unavailable or blocked, use MCP backup:
   • mcp__context7__resolve_library_id
   • mcp__context7__query_docs
3. If both CLI and MCP are unavailable, use API backup with local helper:
   • op run --env-file ~/.codex/.env -- python3 scripts/context7.py search <library_name> --query "<query>"
   • op run --env-file ~/.codex/.env -- python3 scripts/context7.py context <library_id> "<query>" --type md
4. Answer from retrieved docs and label any inference as inference.

Lane B: Skill wizard / CLI

1. Match the user request to the correct ctx7 command and options.
2. For generation flows, default to secure invocation:
   • op run --env-file ~/.codex/.env -- ctx7 skills generate
3. Include setup/auth flows when requested:
   • ctx7 setup, ctx7 login, ctx7 whoami, ctx7 logout
4. Include verification commands (ctx7 skills list ...) and restart guidance.
5. For option/flag uncertainty, use references/context7-skill-wizard.md rather than guessing.

Failure mode

• If no good library match exists, ask for minimum clarification instead of guessing.
• If quota is exhausted, state that explicitly, recommend ctx7 login, and use API backup if available.
• If the request is skill-install/generate and CLI is unavailable, report the blocker explicitly instead of inventing MCP alternatives.

Constraints

• Redact secrets, tokens, credentials, and sensitive data by default.
• Never expose or echo CONTEXT7_API_KEY.
• Treat network access as limited to the Context7 documentation service and returned library metadata.
• Network allowlist for scripts/context7.py: only context7.com and api.context7.com over HTTPS.
• Prefer focused excerpts over full-document dumps.
• Do not invent ctx7 options that are not listed in references/context7-skill-wizard.md.

Validation

• Fail fast: stop at the first failed gate, fix it, then continue.
• Confirm the library id matches the intended ecosystem before using results.
• If results look stale or off-target, refine the query or re-run with narrower scope.
• Cap retrieval attempts per user question:
  • max 3 library resolution attempts
  • max 3 docs retrieval attempts
• For wizard requests, confirm command/flag correctness against references/context7-skill-wizard.md.
• For skills install, enforce repository format /owner/repo.
• See references/contract.yaml and references/evals.yaml for required outputs and eval cases.
• For schema-bound outputs, include schema_version in the response contract.

Examples

• "When the user asks: We just upgraded Next.js and middleware stopped matching. What is the current matcher pattern to exclude static assets and API routes?"
• "User says: Our Supabase RLS policy works locally but fails in prod after a bump. Pull current docs and show the service-role-safe pattern."
• "Can you help me generate a custom skill for OAuth hardening with op run --env-file ~/.codex/.env -- ctx7 skills generate, and explain where it installs?"
• "Can you install every skill from /anthropics/skills for both Cursor and Claude, then show me how to validate the install?"
• "Can you set up Context7 in MCP mode for Claude, and confirm auth is active?"

Anti-patterns

• Guessing API behavior without checking current docs.
• Using outdated versions or deprecated endpoints without calling that out.
• Dumping large doc excerpts instead of answering the user’s actual question.
• Treating a weak library match as authoritative.
• Inventing wizard subcommands/flags without reference validation.

See Also

Topic map: [[product-strategy]]

References and assets

• Open the execution contract: references/contract.yaml
• Open eval coverage and adversarial cases: references/evals.yaml
• Open Context7 CLI wizard command map and options: references/context7-skill-wizard.md
• Open extended strategy and fallback decision notes: references/decision-guidance.md
• Task profile for graph/runtime metadata: references/task-profile.json
• Local helper script: scripts/context7.py
• Skill visual asset: assets/context7.png
• OpenAI Apps metadata and icons: agents/openai.yaml, agents/assets/icon-small.png, agents/assets/icon-large.png

Decision feedback protocol (required):

• If post-run feedback capture is enabled for this runtime, emit a non-blocking post_run_feedback event via request_user_input after result delivery.
• Capture: decision (accepted|partial|rejected|deferred), outcome (good|neutral|bad|unknown), and confidence (high|medium|low).
• Persist with: python3 utilities/skill-builder/scripts/record_skill_feedback.py --skill-path <path/to/SKILL.md> --decision <...> --outcome <...> --confidence <...> --notes "...".
• The recorder tags subject (for example ui, code_review, backend, security) for cross-domain quality analytics.

Gotchas

• ctx7 skills generate requires login and usually works best via op run --env-file ~/.codex/.env -- ctx7 skills generate.
• Docs lookup can fall back to MCP or API, but skill install/generate flows remain CLI-only.
• Skill wizard installs may require agent restart to appear in discovery lists.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.