The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from local markdown files (decisions.md, seed.md, proposal.md) to update a roadmap.md file. This represents a potential indirect prompt injection surface if these files are populated with untrusted data.
Ingestion points: Files located within the docs/threads/ directory.
Boundary markers: The skill does not define explicit delimiters (e.g., XML tags) for the content it reads.
Capability inventory: The agent is permitted to edit roadmap.md and invoke the /emit-pending-decisions tool.
Sanitization: The skill implements logic to detect changes that "alter human intent" and queues them for manual review rather than executing them silently.
[COMMAND_EXECUTION]: The skill uses a custom tool, /emit-pending-decisions, to manage unresolved decisions. This appears to be a structured internal command for project state management rather than an interface for executing arbitrary shell commands.

reconcile-roadmap