The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a shell script scripts/init-brain.sh used to scaffold the project structure. This script executes standard commands (git, mkdir, cat) to create a local .agent/brain/ directory and populate it with markdown templates. These operations are limited to the project workspace and do not involve remote downloads or privilege escalation.
[PROMPT_INJECTION]: The skill establishes a pattern for the agent to ingest external data from the .agent/brain/ directory to influence its output tone and content.
Ingestion points: Instructions in SKILL.md direct the agent to load files like brand-voice.md and audience-profiles.md into the active context.
Boundary markers: None explicitly defined in the provided files to delineate between brand context and task instructions.
Capability inventory: The agent can read files, write to a state file (.agent/state/last-run.json), and execute a local initialization script.
Sanitization: No content sanitization is performed on the ingested markdown files, which is typical for internal project documentation. The risk is minimized as the data source is the project's own repository.

business-brain