| name | setup |
| description | Initialize cognitive-core configuration for this project. Creates cognitive-core.conf with project-specific settings. |
| user-invocable | true |
| allowed-tools | ["Bash","Read","Write","Edit","Glob","Grep"] |
cognitive-core Setup — Project Configuration Wizard
Initialize cognitive-core for the current project by generating a cognitive-core.conf file.
Pre-flight Checks
- Check if
cognitive-core.conf already exists at the project root. If yes, ask the user if they want to reconfigure or keep existing.
- Check if
.claude/hooks/ directory exists (legacy installation). If yes, warn:
"Detected legacy cognitive-core installation in .claude/hooks/. The plugin provides hooks natively — you can safely remove .claude/hooks/, .claude/agents/, and .claude/skills/ directories to avoid duplicate hook execution."
Configuration Steps
Gather the following from the user conversationally. Use sensible defaults based on project analysis.
1. Project Identity
- CC_PROJECT_NAME: Detect from
package.json, pom.xml, Cargo.toml, pyproject.toml, or directory name
- CC_PROJECT_DESCRIPTION: Ask or detect from README
- CC_ORG: Detect from git remote URL
2. Language & Stack
Detect primary language by scanning file extensions. Present finding and ask to confirm:
- CC_LANGUAGE: python, node, react, angular, spring-boot, java, go, rust, perl, csharp
- CC_LINT_COMMAND: Suggest based on language (ruff, eslint, go vet, etc.)
- CC_TEST_COMMAND: Suggest based on language (pytest, jest, go test, etc.)
3. Database
- CC_DATABASE: Detect from dependencies (psycopg2 → postgresql, mysql-connector → mysql, cx_Oracle → oracle) or ask. Default: none
4. Architecture
- CC_ARCHITECTURE: Detect from directory structure (layered, hexagonal, mvc, microservice, monolith). Default: layered
- CC_SRC_ROOT: Detect (src/, lib/, app/)
- CC_TEST_ROOT: Detect (tests/, test/, tests/)
5. Security Level
Explain the three levels and ask:
- minimal: Blocks only critical destructive commands (rm -rf /, DROP DATABASE)
- standard (recommended): Blocks destructive commands + sensitive file reads + untrusted domains
- strict: Blocks everything not explicitly allowed. Best for production/compliance environments.
6. Git
- CC_MAIN_BRANCH: Detect from git (main or master)
- CC_COMMIT_FORMAT: Default to conventional
Output
Generate cognitive-core.conf at the project root using the template format from the cognitive-core framework. Include all sections with comments.
After writing the config, also generate a minimal CLAUDE.md at the project root if one does not exist, containing:
- Project name and description
- Quick reference table (language, test command, lint command, main branch)
- Architecture section
- Key rules section
Language Pack Recommendation
After setup, check if there is a matching language pack plugin available:
- react →
cognitive-core-react
- angular →
cognitive-core-angular
- spring-boot →
cognitive-core-spring-boot
- python →
cognitive-core-python
- perl →
cognitive-core-perl
If a language pack exists, suggest:
"A language pack is available for your stack. Install it with: claude plugin install cognitive-core-<language>@mindcockpit-tools"
Note: Language pack plugins are coming soon (Phase 3). For now, suggest using the legacy install.sh for language-specific features.