The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from an external configuration file, creating an indirect prompt injection surface.\n
Ingestion points: Reads ICP filters and logic parameters from gtm-brief.yaml as described in SKILL.md.\n
Boundary markers: None; the instructions do not include delimiters or warnings to treat the YAML content as untrusted.\n
Capability inventory: The skill generates and writes executable scripts (script.py, script.ts, script.mjs) and executes local shell scripts (scripts/detect_pm.sh, scripts/verify_sdk.sh).\n
Sanitization: No explicit sanitization or validation of the YAML data is performed before it is interpolated into generated code, which could lead to malformed or malicious script generation if the input file is compromised.\n- [COMMAND_EXECUTION]: Runs local shell scripts to determine the appropriate package manager and verify SDK installation.\n
Evidence: Execution of scripts/detect_pm.sh and scripts/verify_sdk.sh documented in SKILL.md.\n- [EXTERNAL_DOWNLOADS]: Installs the blitz-api-py and blitz-api-js libraries via standard package managers.\n
Evidence: Installation commands for uv, pip, bun, and npm are provided in scripts/detect_pm.sh. These packages are legitimate resources associated with the author's brand.\n- [CREDENTIALS_UNSAFE]: References the BLITZ_API_KEY but provides robust guidance on secure handling to prevent exposure.\n
Evidence: Instructs users to use .env files and update .gitignore in references/error-handling.md to ensure secrets are not committed to version control.

blitz-create-script