The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze conversation transcripts, which are considered untrusted inputs. It includes explicit boundary markers and instructions for sub-agents to treat this data as untrusted, ignore any embedded directives within the transcripts, and confine tool usage to referenced context only. This represents a proactive mitigation against indirect prompt injection attacks. Evidence: references/divergent-reviewer.md, references/judgment-reviewer.md, references/tooling-reviewer.md, and references/synthesizer.md.- [SAFE]: The skill instructions enforce safe operational practices, including the requirement for explicit user approval before applying any proposed changes to source repositories and a prohibition against silent patching of the environment. Evidence: SKILL.md.

pstack-reflect