The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external sources, including competitor repositories, released applications, and reverse-engineered packages (ASAR, APK, etc.). This data is then used to inform implementation decisions and code changes.
Ingestion points: Competitor repositories, released binaries, and application packages (found in SKILL.md and references/skill-composition.md).
Boundary markers: The skill instructs the agent to "distinguish assertion from probe," "exclude hidden reasoning," and "treat assistant conclusions as hypotheses" (found in SKILL.md and references/evidence-and-fit.md).
Capability inventory: The agent is authorized to perform repository engineering, implement code changes, and handle deployment/production writes within authorized boundaries (found in SKILL.md and references/implementation-and-acceptance.md).
Sanitization: No specific technical sanitization or filtering of the ingested content is described before it influences implementation logic.
[EXTERNAL_DOWNLOADS]: The skill explicitly directs the agent to reference and potentially load external "specialist" skills from untrusted GitHub repositories for reverse engineering tasks.
Evidence: References to https://github.com/morluto/rea and https://github.com/zhaoxuya520/reverse-skill in references/skill-composition.md.
Context: The skill warns that "Skill-only integration does not authorize configuration writes, downloads or engine installation" and requires "separate applicable setup authorization," which reduces the immediate risk but does not eliminate the dependency on untrusted external instructions.

app-evolver