The Agent Skills Directory

[DYNAMIC_EXECUTION]: The audit procedure defined in docs/audit.md generates a shell script at runtime by writing to /tmp/slowed.sh and then executes it by assigning it to the EDITOR environment variable during jj command execution.
[PRIVILEGE_ESCALATION]: The docs/audit.md file instructs the agent to use chmod +x on the dynamically created /tmp/slowed.sh script to grant it execution permissions.
[COMMAND_EXECUTION]: The skill relies heavily on the agent executing shell commands (jj log, jj squash, jj git push, etc.) to interact with the version control system and manage the repository state.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and process external, potentially untrusted data from repository logs, diffs, and commit messages.
Ingestion points: Data enters the agent's context through commands like jj log, jj diff, and reading file contents within the .jj/ or .git/ directories.
Boundary markers: The instructions do not define specific delimiters or "ignore" instructions for content found within the repository data.
Capability inventory: The agent has the capability to write to the file system and perform network operations via jj git push based on its interpretation of the repository state.
Sanitization: There is no evidence of sanitization or validation performed on the ingested repository content before it is used to influence further agent actions.

jj