The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes a variety of external artifacts to generate implementation plans, which presents a surface for indirect prompt injection.
Ingestion points: As defined in references/artifact-inputs.md, the skill reads tickets, task statements, research, design discussions, product requirements, technical designs, and user feedback.
Capability inventory: The skill has the capability to read and write files within the repository and execute shell and Git commands to manage branches and verify implementation details.
Boundary markers: While the skill establishes clear precedence rules for resolving conflicts between artifacts (e.g., structure outlines vs. technical designs), it lacks explicit boundary markers or instructions to treat data within those artifacts as untrusted instructions.
Sanitization: No evidence of content sanitization or instruction filtering is present for the data ingested from task artifacts.
[COMMAND_EXECUTION]: The skill involves the execution of local shell and Git commands to inspect the repository and manage task branches.
Evidence: The instructions in SKILL.md and references/handoff-routing.md require the use of ls -La for directory inventory, and Git commands including git rev-parse, git branch, git status, and git switch to gather environment metadata and prepare the workspace for implementation.

create-plan