The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to perform file operations (mkdir, cp) and initialize project-local subsystems. It executes initialization scripts for specific domains such as appsec-sdk.sh, qa-sdk.sh, and uiux-sdk.sh. These scripts are located in the user's local ~/.claude/ configuration directory and are part of the skill's expected operating environment.
[EXTERNAL_DOWNLOADS]: Through a dedicated provision-tools.js script, the skill can install project-specific development tools using standard package managers including npm, pip, go, cargo, brew, and apt. The skill explicitly gates offensive security tools (tier roe) and requires manual user activation for such installations.
[DYNAMIC_EXECUTION]: The skill dynamically generates and executes commands to register project hooks in .claude/settings.json. The selection of which hooks to install is determined at runtime based on the tech stack and signals detected during the project scanning phase.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data (source code, README files, and manifests) to build a signal_vector for configuration matching.
Ingestion points: Project source code, package.json, README.md, and deployment configurations are read via Glob, Grep, and Read tools during Step 1 (SCAN).
Boundary markers: The skill normalizes project data into a structured signal_vector schema defined in templates/SCAN-VECTOR-SCHEMA.md before processing it.
Capability inventory: The skill possesses Bash, Write, and Edit capabilities to modify the project environment.
Sanitization: All proposed changes, including file writes and tool installations, are presented to the user for explicit confirmation in the PROPOSE step before execution.

claude-env-bootstrap