The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill defines a 'spike' process where the agent is instructed to write and execute small, throwaway runnable experiments in a specific directory (.aida-spike/) within the task worktree to answer design questions that lack existing documentation.
[INDIRECT_PROMPT_INJECTION]: The research process involves ingesting and summarizing data from external sources via search agents (outward-searcher, catalog-identifier). This introduces a vulnerability where malicious content within those external sources could attempt to influence the agent's summaries or the creation of experimental 'spike' code.
Ingestion points: Data returned by search agents and files located in the task's inputs/ folder are processed into findings.
Boundary markers: The skill requires findings to be recorded with specific metadata (source, date, text), which provides structural delimitation but lacks explicit instructions for the model to ignore embedded commands within the findings text.
Capability inventory: The skill possesses the ability to write findings to the file system, generate and execute arbitrary code via the 'spike' mechanism, and dispatch additional agents.
Sanitization: The skill utilizes the jq utility to handle JSON data, ensuring that finding text is correctly escaped within the data structure, though it does not perform semantic filtering of the content itself.
[COMMAND_EXECUTION]: The skill uses local Bash scripts (research-actions.sh, playbook-actions.sh, project-actions.sh, etc.) to manage the task lifecycle, including recording research data and performing coverage checks. Access to these scripts is explicitly restricted via the allowed-tools configuration in the skill's metadata.

research