The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs several local command executions, including Git operations (git checkout, git pull, git remote) and status updates using the marsh CLI tool or jq for atomic file writes.
[INDIRECT_PROMPT_INJECTION]: The skill reads the issueBody field from a local state file and displays it to the agent, creating a surface where untrusted data could influence agent behavior.
Ingestion points: SKILL.md (Step 11) reads data from the local JSON file located at ${MARSHROOM_STATE:-~/.config/marshroom/state.json}.
Boundary markers: The skill does not implement delimiters or safety instructions when displaying the retrieved issue body to the agent context.
Capability inventory: The skill has the ability to execute shell commands (git, marsh, jq) and perform file system writes (updating the state JSON file).
Sanitization: There is no evidence of validation or sanitization for the content retrieved from the issueBody field before it is processed by the agent.

start-issue