The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to autonomously execute shell commands (verify and guard commands) provided during the setup phase. The instructions explicitly direct the agent to 'Never stop. Never ask' and to repeat the execution indefinitely, which increases the potential impact of any misconfigured or malicious commands.
[INDIRECT_PROMPT_INJECTION]: The agent is instructed to read project files and experiment logs to 'ideate' new changes. This creates a surface for indirect prompt injection where malicious instructions hidden in the code or documentation could influence the agent's modifications or command executions.
Ingestion points: During the 'Review' phase, the agent reads in-scope project files, the autoresearch-results.tsv log file, and git branch names.
Boundary markers: The skill lacks explicit instructions for the agent to ignore or delimit instructions found within the project files it analyzes.
Capability inventory: The skill utilizes shell command execution (via verify/guard commands), file system modification (editing source code and logs), and network communication (git push to remote repositories).
Sanitization: No sanitization or validation mechanisms are defined for the data ingested from the project environment before it is used to inform the agent's decision-making process.

autoresearch