The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external plan files which serves as a potential attack surface for indirect prompt injection.
Ingestion points: The skill reads tasks.md and docs/agents/project.md to establish the execution context and task capsules.
Boundary markers: The skill generates a task-N-brief.md for each task, which includes a content hash of the original plan to ensure consistency.
Capability inventory: The agent is authorized to use git for version control and perform file system writes to implement tasks as defined in the plan.
Sanitization: The 'Iron Law' explicitly instructs the agent to 'Stop and ask' if the plan is ambiguous or contradictory, preventing the agent from guessing or following potentially malicious hidden instructions in the data.
[COMMAND_EXECUTION]: The skill utilizes standard version control commands for development workflows.
Evidence: The skill uses git rev-parse HEAD and git log to track progress and verify the state of the repository.
Context: These commands are limited to local repository metadata and do not involve administrative privileges or network exfiltration.

build-inline