The Agent Skills Directory

[SAFE]: The skill is focused on improving transparency and auditability. No malicious patterns such as data exfiltration, obfuscation, or unauthorized command execution were found.
[COMMAND_EXECUTION]: The skill includes a local utility script scripts/log.sh. This script is used to append formatted text to a local file. It uses standard bash utilities and implements safe handling of input variables to prevent unexpected behavior.
[INDIRECT_PROMPT_INJECTION]: While the skill processes data that could be influenced by external sources (e.g., commit messages or PR titles as evidence), it incorporates multiple layers of protection.
Ingestion points: Untrusted data enters the agent context through the 'evidence' and 'decision' fields described in SKILL.md.
Boundary markers: The use of a TSV schema and a dedicated logging script enforces data boundaries.
Capability inventory: The skill uses scripts/log.sh for local file writes.
Sanitization: The clean() function in scripts/log.sh strips newline/tab characters and prefixes characters like '=', '+', '-', and '@' with a single quote to mitigate CSV/spreadsheet injection attacks.
[DATA_EXPOSURE]: The skill writes information to the local project directory (e.g., 'decisions.tsv'). It does not attempt to access sensitive system files or credentials.

show-me-your-work