The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for auditing decision logs and transcript data which may contain untrusted input. * Ingestion points: The skill reads local files such as decisions.tsv and files within the agent-transcripts/ directory. * Boundary markers: The instructions provide explicit guidance to avoid reading data outside the current project scope (e.g., ~/.cursor/projects/*/). * Capability inventory: The skill is capable of reading and writing local text files. * Sanitization: The provided scripts/log.sh includes a clean() function that proactively mitigates CSV injection (Formula Injection) by stripping tabs and newlines and prefixing potential formula triggers (characters like =, +, -, and @) with a single quote before writing to the log.
[COMMAND_EXECUTION]: The skill includes a local utility script used to manage the logging process. * Evidence: The script scripts/log.sh is a standard bash utility that uses safe coding practices, such as set -euo pipefail and input cleaning, to append structured data to a local file. It does not execute remote code or perform unsafe network operations.

show-me-your-work