The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes task descriptions and existing task lists which are often sourced from external files or user input. This creates an attack surface where instructions embedded within a task list could attempt to override agent behavior during the processing or commit phases.
Ingestion points: SKILL.md workflow step 2 describes reading existing host plans or task lists into the agent context.
Boundary markers: The skill defines an 'Execution boundary' in SKILL.md that explicitly separates read-only analysis and preview phases from modification and deployment phases, reducing the risk of accidental execution.
Capability inventory: The skill utilizes host-provided capabilities for file searching, CLI operations, and browser access, and includes a local Python utility (profile_store.py) for profile management.
Sanitization: While the skill requires explicit parsing of project context, it does not detail specific sanitization or filtering logic for the text content of tasks being processed.
[COMMAND_EXECUTION]: The skill is designed to interact with the host environment's CLI tools to perform task updates, commits, and deployments. The execution is guided by natural language instructions, relying on the host agent's native tool-calling safety mechanisms.
[SAFE]: The provided Python script scripts/profile_store.py implements secure profile management. It uses atomic writes for data integrity and includes a hardcoded blacklist (SENSITIVE_PARTS) to explicitly block the storage of secrets, tokens, and credentials in the durable profile records.

lov-add-task