The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes issue bodies and comments from GitHub or Linear, which are external data sources. This creates a surface for indirect prompt injection if a malicious user provides instructions in a comment.
Ingestion points: SKILL.md Step 3 reads issue content and comments via gh issue view and Linear MCP tools.
Boundary markers: While not explicitly using XML tags for the LLM, the skill instructions implement a detailed 'Author Authority' framework to delineate untrusted text.
Capability inventory: The skill possesses the ability to modify issue content (gh issue edit), search repositories (gh search issues), and view commit histories (git log).
Sanitization: The REFERENCE.md file defines strict rules for authorized authors (admins, maintainers, trusted bots). Text from unauthorized users is treated as context and cannot override the agent's instructions. Any direct addressing of the agent in unauthorized text is flagged for human review.
[COMMAND_EXECUTION]: The skill executes a bundled shell script, templates/resolve-config.sh, to merge configuration files. This execution is performed using the sh command and is restricted to the local environment and the specific config file.
[EXTERNAL_DOWNLOADS]: The skill references an external JSON schema at https://raw.githubusercontent.com/TitusKirch/skills/main/tituskirch-skills.schema.json. This is an official resource from the skill's author used for validation and does not involve the execution of remote code.

refine-issue