The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from backlog items, handoff notes, and plan artifacts located in .rota/plans/*.md or external issue trackers.
Ingestion points: plan-artifact.md loads plans using rota plan show, and no-argument-mode.md ingest handoff notes via rota status handoff.
Boundary markers: The skill uses a template in dispatch-brief.md to structure subagent instructions, but it does not employ robust delimiters or explicit 'ignore instructions' warnings to prevent the agent from obeying malicious content within the plan artifacts.
Capability inventory: The main agent session has the capability to execute shell commands (via rota proof record), perform git commits, and manage subagents that write files.
Sanitization: There is no evidence of sanitization or filtering for the 'verify step' strings retrieved from plans before they are passed to the shell for execution.
[DYNAMIC_EXECUTION]: The skill uses rota proof record in Step 7 of SKILL.md to execute 'verify steps' which are dynamically retrieved from the plan artifacts. This mechanism allows the execution of shell commands defined in data rather than the skill's own code.
[COMMAND_EXECUTION]: The skill frequently executes CLI commands, including standard tools like git and godot, as well as a suite of project-specific rota utilities used for state management and task tracking.

rota-work