The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a local script scripts/log.sh to maintain a decision log. This script executes an embedded Python helper that implements high-standard security controls, including path validation to ensure writes stay within the workspace, prevention of symlink and hardlink attacks, and a built-in secret scanner that blocks the logging of sensitive credentials (API keys, tokens, etc.).
[EXTERNAL_DOWNLOADS]: The MIRROR.md file documents the skill's origin as the github.com/cursor/plugins repository. This is a well-known source and the reference is provided for supply chain transparency rather than dynamic runtime execution from untrusted sources.
[INDIRECT_PROMPT_INJECTION]: The skill involves reading and writing to a TSV log file, which represents a potential surface for indirect prompt injection if malicious data is logged.
Ingestion points: The agent is instructed to read the log file during the audit and review phases to verify work history.
Boundary markers: The log uses a structured TSV format, which provides clear separation between different data fields.
Capability inventory: The skill has the capability to append to local files via the logging script.
Sanitization: The logging script actively sanitizes input by removing delimiters (tabs/newlines) and escaping formula-triggering characters (such as =, +, -, and @) with a single quote. This mitigation effectively prevents both spreadsheet-style injection and reduces the risk of the agent misinterpreting logged data as commands.

cursor-show-me-your-work