The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a bash script to create a secure temporary directory in /tmp. The script employs security checks to prevent symlink attacks and ensures directory ownership by the current user using commands like id, mkdir, and chmod.
[DATA_EXFILTRATION]: The skill can publish generated explanations to an external service (ht-ml.app). The instructions mandate that the agent must warn the user about the public nature of the service and obtain explicit confirmation before any transmission occurs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository to generate its outputs. 1. Ingestion points: Git commit logs, source files, and pull request data via CLI tools as specified in references/orchestration.md and references/agents/work-recap-scout.md. 2. Boundary markers: The skill relies on grounding instructions to distinguish evidence from inference rather than technical delimiters. 3. Capability inventory: Command execution for directory creation, file writes to /tmp, and network operations for publishing via tools or other skills. 4. Sanitization: The agent is instructed to check every factual claim against its source, remove unsupported claims, and label unverified information derived from model knowledge.

ce-explain