The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to process data sources that frequently contain sensitive information, including shell history (~/.zsh_history), git commit messages, and AI chat histories.
Evidence: The instructions explicitly acknowledge this risk: "History can contain credentials. Filter before printing tool output, never emit raw histories, and omit secrets and unnecessary personal or customer data from the post."
Mitigation: The skill includes a clear safety protocol directing the agent to warn the user if a credential is found without revealing its value in the final output.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources that could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The agent reads GitHub PR descriptions, git commit messages, and local file changes (SKILL.md).
Boundary markers: No specific delimiters (like XML tags or triple quotes) are defined to separate the untrusted content from the system instructions.
Capability inventory: The agent uses the gh CLI for network reads and pbcopy to write to the macOS clipboard (SKILL.md).
Sanitization: The instructions focus on filtering credentials rather than sanitizing the content for embedded instructions.

telegram-daily-log