The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local bash script scripts/log.sh to append entries to a TSV log file. It also invokes an external utility ../rex-subagents/scripts/sessions to retrieve and read session transcripts for auditing purposes.
[INDIRECT_PROMPT_INJECTION]: During the audit phase, the agent is instructed to read the session transcript, which includes tool outputs. These outputs may contain data from external sources (e.g., web pages, files) that could include malicious instructions intended to manipulate the agent's behavior during the log verification process.
Ingestion points: Transcript data retrieved via ../rex-subagents/scripts/sessions read <id> --tools.
Boundary markers: The instructions do not specify explicit delimiters or 'ignore' instructions when processing transcript content.
Capability inventory: The skill has the ability to write to the filesystem (via log.sh), execute scripts, and spawn sub-workers.
Sanitization: The scripts/log.sh script includes a clean() function that strips control characters and prefixes spreadsheet formula triggers (=, +, -, @) with a single quote to mitigate CSV injection risks.
[DYNAMIC_EXECUTION]: The skill provides instructions for the agent to dynamically locate and follow other instruction files (skills and principles) residing in neighboring directories (e.g., ../<name>/SKILL.md) based on names provided in the context.

show-me-your-work