The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an autonomous learning pipeline that clones external repositories and extracts patterns to populate procedural memory (.claude/rules/learned/). This creates a surface for indirect prompt injection, where an attacker could embed malicious instructions in a public repository to influence the agent's future behavior.\n
Ingestion points: Untrusted data from external GitHub repositories enters the agent context via curator-ingest.sh and curator-learn.sh.\n
Boundary markers: No explicit delimiters or boundary markers are described to isolate or ignore embedded instructions within the processed repository content.\n
Capability inventory: The skill uses Bash, Write, Read, WebSearch, and WebFetch to download repository data and modify local rule files.\n
Sanitization: There is no mention of sanitization, validation, or filtering of the content extracted from external repositories before it is interpolated into the agent's memory system.\n- [COMMAND_EXECUTION]: The skill relies on executing multiple orchestration scripts (curator-discovery.sh, curator-scoring.sh, curator-rank.sh, curator-ingest.sh, curator-approve.sh, curator-learn.sh) to manage the repository processing pipeline.\n- [EXTERNAL_DOWNLOADS]: The skill fetches repository metadata and clones source code from GitHub to facilitate its learning and discovery functions.

curator