The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various Python scripts located within the project's .trellis/scripts/ directory to manage task lifecycles, developer identity, and session context (e.g., python3 ./.trellis/scripts/task.py current).
[DYNAMIC_EXECUTION]: The skill architecture relies on 'lifecycle hooks' defined in .trellis/config.yaml and platform-specific settings (like .claude/settings.json). These hooks allow the execution of arbitrary shell commands and scripts at specific project events (e.g., after_create, after_finish), which the agent is encouraged to modify to customize behavior.
[INDIRECT_PROMPT_INJECTION]: The agent is designed to ingest and act upon data from untrusted local files, including Product Requirement Documents (prd.md), technical designs (info.md), and engineering specifications (.trellis/spec/).
Ingestion points: The agent reads content from the .trellis/tasks/ and .trellis/spec/ directories to determine implementation steps and quality rules.
Boundary markers: The system uses specific file manifests (implement.jsonl, check.jsonl) to scope context, though it does not explicitly define safety delimiters for the content within those files.
Capability inventory: The agent has the capability to write files, execute shell commands, and modify the project's internal automation scripts.
Sanitization: The skill does not describe specific sanitization steps for content read from project files before it is processed by the AI.
[PERSISTENCE]: The use of lifecycle hooks in config.yaml and platform hooks in directories like .claude/hooks/ provides a mechanism for code to be executed persistently across different development sessions based on project events.

trellis-meta