Appliquer une compétence réutilisable au projet

---

description: Apply and integrate a reusable skill into the current project with guided analysis and approval. Everything runs inside Claude - no terminal commands needed.

Model Guidance

This command does implementation work. Use claude-sonnet-4-6 for this session.

Execution against a defined plan - this is where call volume lives and where savings compound.

---

User Input

Telemetry: Run: polaris telemetry record skill --feature <slug> --phase start --agent claude

$ARGUMENTS

You MUST consider the user input before proceeding (if not empty). Input should be a skill name (e.g., iam, mcp-server, studio, devops, workflow, appcentral-shell).

Hard Rules

1. NO POLARIS CLI COMMANDS -- only the Step 1 bash command below. Use Read/Glob/Grep for everything else.
2. NO SUBAGENTS for skill analysis -- all skill analysis done directly with Read, Glob, Grep tools. (Subagents are used in implement context only, not here.)

Steps

1. Discover Skills

The Python package is specify_cli (NOT polaris). Locate its install path without invoking python directly - on Windows, python and python3 may resolve to the Microsoft Store stub even when Python is installed, which produces a confusing "not installed" failure.

1a. Find the install location. Try in order, stopping at the first one that returns a real path:

pip show specify_cli 2>&1 | grep -E "^Location:"

If pip returns nothing or "not found":

pipx list --short 2>&1 | grep -E "(^|\s)specify_cli(\s|$)"

If both fail with "not found" (not just a missing pip / Store-stub error), STOP and tell the user specify_cli is not installed in any reachable Python environment, and suggest pip install specify_cli (or the team's private install command).

If pip itself fails with a "Python was not found" / Microsoft Store / exit-code-49 error, the issue is the Windows Python alias, not a missing package. Tell the user to either disable the Microsoft Store Python alias (Settings > Apps > Advanced app settings > App execution aliases) or run pip from the directory of their real Python install. Do NOT conclude that specify_cli is missing.

1b. Compute the base path. Parse the Location: value from pip show. The skills base is <Location>/specify_cli/skills/builtin. Store this as BASE_PATH.

1c. Enumerate skills with Glob (do not list the directory via bash). Pattern: <BASE_PATH>/*/SKILL.md. The unique parent directory names are the available skill names.

If no skill name was provided in $ARGUMENTS, show the discovered skill names and ask the user which one to apply.

2. Analyze Project

Scan with Glob/Grep (parallel where possible, NO bash):

• Entry points: **/main.py, **/app.py, **/index.ts, **/Program.cs
• Package manager: poetry.lock, uv.lock, yarn.lock, pnpm-lock.yaml, pyproject.toml, package.json
• Env files: .env, .env.example -- Read any found
• Structure: Check frontend/, client/, web/ for package.json and .git (separate repos?)
• Active skills: .polaris/skills/*.md
• Existing skill code (Grep for skill-specific patterns):
  • iam: Depends\(get_current_user\)|require_auth|JWTBearer|passport\.authenticate
  • studio: APTEAN_STUDIO|AzureOpenAI|studio_client
  • devops: Glob .github/workflows/*.yml, Dockerfile
  • workflow: APTEAN_WORKFLOW|workflow_client
  • mcp-server: FastMCP|McpServer|@mcp\.tool
  • appcentral-shell: useAppCentralLocale|useAppCentralTheme|APPCENTRAL_LOCALE_CHANGE|APPCENTRAL_THEME_CHANGE|AppCentralShellProvider
• Routes: @app\.(get|post|put|delete)|@router\.|app\.use\(
• Frontend code: Glob frontend/src/**/*.{ts,tsx,js}, src/**/*.{ts,tsx,jsx}, static/**/*.{html,js}
  • Look for: field access patterns on API responses, auth patterns, API call patterns (fetch/axios)
• Tests: Glob tests/**/*.py or **/*.test.ts, Read 1-2

Determine: framework, package manager, project structure (monolith/separate-folders/separate-repos), frontend root, existing skill code, consumer field expectations, frontend auth gaps.

3. Read SKILL.md, Config Questions, Deploy Guardrail

3a. Read <BASE_PATH>/<skill-name>/SKILL.md. Parse config: frontmatter entries (key, type, prompt, default, options).

3b. Pre-fill answers from Step 2 analysis (e.g., detected helm/ -> DEPLOY_TARGET=helm, APTEAN_IAM_* env vars -> AUTH_PROVIDER=keycloak). Tell user what was detected.

3c. Ask remaining config questions interactively. For iam: ALWAYS ask identity provider -- never silently default to Entra. Company default is Aptean IAM (Keycloak). If user presses Enter, use defaults.

3d. Read <BASE_PATH>/<skill-name>/guardrail.md, replace {{PROJECT_NAME}} with project dir name, write to .polaris/skills/<skill-name>.md.

4. Read Guardrail

Read the deployed guardrail. Its Patterns, Anti-Patterns, Testing Rules, and File Conventions are non-negotiable constraints. Config answers determine which path to follow.

5. Cross-Skill Awareness

Read .polaris/skills/ for active skills. Note interactions (e.g., IAM+Studio share tokens, DevOps CI should test all skills).

6. Present Integration Plan -- WAIT FOR APPROVAL

Do NOT modify files until approved. Present: framework, package manager, config, project structure, frontend root, consumer field expectations, existing code found, active skill interactions, files to create, files to modify, frontend changes, auto-actions (install deps, run tests, setup .env).

Wait for: approve all / skip specific files / cancel / questions.

7. Execute Integration

Apply config-driven implementation following guardrail patterns:

• Wire imports (match project style, no duplicates)
• Register middleware (correct order, integrate with existing)
• Protect routes per guardrail spec (skip already-protected)
• Auto-wire consumer fields: match existing frontend field names (e.g., user.username not preferred_username)
• Environment variables: append to .env.example, copy to .env if missing
• Tests: follow project patterns, include positive + negative cases
• Frontend: follow guardrail's Frontend section; inject auth headers, wire UI components; note separate-repo commits

8. Install Dependencies

Use detected package manager (pip/poetry add/uv add/npm/yarn/pnpm/dotnet restore). Stop on failure.

9. Validate and Test

Syntax check modified files, run skill tests, fix failures, verify guardrail compliance (no anti-patterns).

10. Summary

Past-tense summary: config used, guardrail deployed, files created/modified, deps installed, test results, brownfield/frontend notes. Only manual step: fill .env credentials.

Context: $ARGUMENTS

Telemetry: Run: polaris telemetry record skill --feature <slug> --phase complete --agent claude