The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs agents to read and audit decisions.tsv, which contains externally-sourced data such as command outputs, test results, and evidence pointers. This creates a surface where malicious instructions embedded in a command's output could be interpreted by the auditing subagent during the trail review process.
Ingestion points: The decisions.tsv log file is read during the "Audit the trail" and "Reading the trail" phases defined in SKILL.md.
Boundary markers: The log uses TSV (Tab-Separated Values) format to structure data, though it lacks explicit natural language boundary markers to isolate data from instructions for the auditing subagent.
Capability inventory: The skill includes the ability to write to the local filesystem using scripts/log.sh and mentions the use of independent subagents for inspection.
Sanitization: The scripts/log.sh script performs structural sanitization by removing tabs and newlines from log cells and protects against CSV/spreadsheet formula injection by prefixing characters like =, +, -, and @ with a single quote.
[SAFE]: The skill incorporates materials adapted from the Cursor plugins repository, which is an established service for AI development tooling. These components are used for their intended purpose of providing decision-logging functionality.

show-me-your-work