The Agent Skills Directory

[SAFE]: The skill implements a robust verification strategy, requiring the agent to call get_project to confirm the existence and metadata of a target project before any deletion action is initiated.
[SAFE]: It mandates a strict 'Preview and Approve' loop, where the user must explicitly approve the precise resource ID (projects/{project}) after seeing the project title, preventing accidental or guessed deletions.
[SAFE]: The skill forbids automatic retries on tool timeouts or network errors, requiring the agent to perform a read-only check using list_projects to determine the actual state of the project rather than blindly repeating a destructive operation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project metadata which could theoretically contain instructions intended to influence the agent.
Ingestion points: Project titles and metadata fetched via get_project and list_projects (referenced in SKILL.md and workflow.md).
Boundary markers: The skill requires an explicit preview step where all metadata is presented to the user for manual verification before the deletion capability is unlocked.
Capability inventory: Access to get_project, delete_project, and list_projects tools.
Sanitization: Human-in-the-loop review of the previewed project data acts as a verification barrier.

stitch-delete-project