The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs git worktree commands to create isolated environments for background processing, ensuring that code analysis tasks do not impact the user's primary working directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests conversation context to generate task briefs for a background agent, representing a potential injection surface. Ingestion points: The agent captures the user's verbatim input and resolves references from the active conversation transcript (SKILL.md, Step 3). Boundary markers: Task briefs are limited to 5 lines and environment fingerprints to 10 lines to minimize context contamination. Capability inventory: The system allows writing to the ~/side-quests/ directory and executing git commands within restricted worktrees. Sanitization: A 'Draft-only' rule explicitly prohibits the background agent from performing actions with side effects, such as sending communications or modifying the main repository.
[DATA_EXPOSURE_EXFILTRATION]: The skill accesses local environment metadata including the current working directory and git branch to provide context for tasks. This data, along with the task ledger, is stored locally in ~/side-quests/ and is not transmitted over the network.

side-quest