The Agent Skills Directory

[SAFE]: No security vulnerabilities, malicious patterns, or unauthorized behaviors were detected across the skill's instructions, schemas, or scripts.
[DATA_EXFILTRATION]: The skill is designed to process and filter sensitive memory record files, specifically patterns.json and gotchas.json. Analysis of the code confirms that these files are accessed only for local deduplication purposes, with no evidence of network exfiltration or unauthorized access.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes memory records that may contain untrusted data. However, the risk is negligible as the skill is not user-invocable (user_invocable: false), operates as a background helper for a specific agent, and performs purely structural data transformations (deduplication).
Ingestion points: Input via schemas/input.schema.json and existing memory files (patterns.json, gotchas.json).
Boundary markers: None explicitly defined in the provided wrapper script.
Capability inventory: Limited to data deduplication and statistics reporting; no file-writing or network capabilities are present in the provided scripts.
Sanitization: Not present in the wrapper, as logic is delegated to the primary memory compression module.

token-saver-memory-dedup