The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources (such as repository code and user requests) to generate implementation plans and delegate tasks. This creates a surface for indirect prompt injection where malicious instructions embedded in a target repository could influence the agent's planning or subsequent execution steps.
Ingestion points: The workflow described in SKILL.md and references/03-deep-planning-interview.md explicitly gathers context from user requests, repository code, and local policy files to define scope and behavior.
Boundary markers: The skill incorporates several mitigation patterns, including 'Rubber Duck Review' (isolated read-only review in references/02-plan-review.md) and a requirement to validate evidence before accepting task completion (SKILL.md). However, it lacks explicit delimiting or instruction-ignoring markers when interpolating untrusted request data into the Business and Technical Plan templates.
Capability inventory: The skill possesses significant capabilities, including writing plan files, delegating tasks to subagents via itsol-subagent-workflow, and executing repository-defined verification commands as part of the TDD (Test-Driven Development) process described in references/05-technical-plan.md.
Sanitization: There is no evidence of automated sanitization, escaping, or filtering of external content before it is processed or used to generate the Task Breakdown for execution.

itsol-functional-planning