The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from existing tracker issues, project templates, and chat conversations to draft or edit issues. This surface is mitigated by instructions to quote reporter feedback directly and a mandatory verification process for all links and attachments.
Ingestion points: The skill reads neighboring issues, parents, and templates in SKILL.md (Normal Path) and chat messages in the 'Bug or request' section.
Boundary markers: The 'State The Claim, Justify Every Link' section (line 122) requires the agent to validate all external references before posting.
Capability inventory: The skill interacts with external trackers (GitHub, Linear, etc.) via dedicated CLI tools and MCP servers to create, modify, and organize issues.
Sanitization: Instructions in SKILL.md (lines 102 and 155) explicitly mandate cutting confidential data from quotes and excluding secrets, credentials, and personal information from tracker records.
[DATA_EXPOSURE]: The skill provides defensive instructions to prevent accidental exposure of local data. It explicitly forbids including paths from the author's local machine or private servers and requires documenting access conditions (e.g., VPN requirements) for shared internal links.

write-issue