The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external agent prompts and configuration files to identify dependencies and update routing logic.
Ingestion points: Phase 1 and Phase 4 read data from files in ~/code/prompts/agents/, ~/code/prompts/skills-lock.json, and .claude-plugin/plugin.json.
Boundary markers: No explicit boundary markers or instructions to disregard embedded commands are used when reading these files.
Capability inventory: The skill has high-impact capabilities including file deletion (rm), git repository modification (git push), and infrastructure management via the clawnet CLI.
Sanitization: No sanitization or validation of the content within the processed markdown or JSON files is performed.
[COMMAND_EXECUTION]: The skill utilizes several system commands to execute its primary purpose of agent removal.
Evidence: Uses rm to delete agent files and avatars, and git commands (add, commit, push) to publish changes. These operations are restricted to specific development paths (~/code/prompts/).
Infrastructure Control: Uses a custom CLI (clawnet) to stop sandboxes and delete bot identities, which is consistent with the skill's documented administrative role.

agent-decommissioning