The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data that could contain malicious instructions designed to subvert the agent's behavior.
Ingestion points: The skill reads README.md and other documentation files from the target repository in scripts/analyze_project.py.
Boundary markers: Prompt templates in templates/prompts/ (e.g., 01-overview-prompt.md) interpolate {project_info} directly into the prompt without delimiters or clear instructions for the AI to ignore embedded commands.
Capability inventory: The agent has the ability to execute shell commands (git, cp, find), write files, and generate images via external scripts.
Sanitization: No sanitization or filtering is performed on the content of the README files before they are included in the prompt context.
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute system commands with user-provided inputs.
In scripts/analyze_project.py, the source (GitHub URL or local path) and token are passed to git clone and cp commands. While passed as a list, maliciously crafted arguments (e.g., strings starting with dashes) could potentially influence command behavior.
In scripts/convert_to_word.py, the skill executes pandoc to transform Markdown content.
[CREDENTIALS_UNSAFE]: The skill explicitly asks users for GitHub Personal Access Tokens to access private repositories. Additionally, scripts/main.py attempts to read an .env file from a hardcoded path belonging to a different skill (~/clawd/skills/nano-banana-pro/.env) to load environment variables, which is a cross-skill data access concern.
[EXTERNAL_DOWNLOADS]: The skill clones remote Git repositories to a local temporary directory (/tmp/tech-analysis-repo) for analysis. This involves downloading arbitrary code from external servers provided by the user.

tech-analysis-reporter