The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads project-specific files within the workspace (e.g., in methods/, code/, and results/ directories) to verify audit artifacts. It does not access sensitive system paths (like .ssh or .aws directories) and contains no network functions to transmit data externally.
[COMMAND_EXECUTION]: No shell commands, subprocess spawning, or arbitrary command execution patterns were found. The skill's operations are limited to reading project files and writing an audit report to paper/audits/completeness_audit.md.
[INDIRECT_PROMPT_INJECTION]: The skill ingests markdown and JSON files produced by other tools to verify their completeness. While this represents a data ingestion surface, the risk is minimal because the skill uses specific mechanical checks (e.g., counting checkmarks, verifying frontmatter keys) and lacks dangerous capabilities like network access or code execution that could be exploited via injection.
Ingestion points: Reads files in methods/, code/, results/, robustness/, paper/, and workspace/.
Boundary markers: Expects structured markdown tables and YAML frontmatter.
Capability inventory: Limited to reading files and writing a local audit report; no execution or network capabilities.
Sanitization: Employs mechanical pattern matching and timestamp comparisons rather than open-ended natural language interpretation of instructions within the files.

completeness-auditor