The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository's existing task and backlog documentation during the migration phase. While the subagents are constrained by explicit scope boundaries and exclusion lists, malicious instructions embedded within pre-existing project documents could attempt to influence the agent's behavior during the reformatting process.
Ingestion points: The references/06-doc-sweep.md subagent is instructed to read legacy documentation identified in the repository analysis phase (DOCS inventory) such as TODO.md or ROADMAP.md files.
Boundary markers: Prompts in references/01-analysis.md and references/06-doc-sweep.md utilize structured instruction fields (GOAL, ROLE, SCOPE, CONTEXT, CONSUMER, DONE) and explicitly define out-of-bounds directories, specifically source code paths and core plugin configuration files.
Capability inventory: The skill utilizes Write, Edit, and Bash tools for local file system management, and the Agent tool to spawn sub-tasks for repo analysis and content migration.
Sanitization: Findings from the initial analysis phase (such as domains, exclusions, and the documentation migration inventory) are presented to the user for explicit confirmation via the AskUserQuestion tool before any file modifications or document sweeps occur.

task-board-setup