The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements dynamic loading of instructions by resolving relative paths to other skills, specifically ../<name>/SKILL.md and ../principle-<name>/SKILL.md. This allows the agent to execute instructions from files determined at runtime.
[DYNAMIC_EXECUTION]: The skill uses the subagent spawn tool to execute workers based on model definitions retrieved from the local configuration file ~/.config/pstack/pstack-models.md.
[COMMAND_EXECUTION]: The skill involves shell-level operations for sub-agent management and repository handling, including the use of subagent spawn and git commands such as git worktree and git commit. It also reads task state from local files located at ~/.local/state/pstack/checklists/.
[INDIRECT_PROMPT_INJECTION]: The aggregation phase (Phase C) involves reading terminal results from sub-workers. This creates an attack surface where worker reports, potentially derived from untrusted external sources during 'exploration' or 'coverage' tasks, could contain malicious instructions. The instructions provide mitigation for this risk by directing the agent to consolidate results into a report table rather than providing raw worker dumps.

swarm