The Agent Skills Directory

[SAFE]: The skill implements a robust architecture for 'Persistent Memory' that prioritizes user control and privacy. All operations (write, read, consolidate, forget) are governed by detailed pipelines with explicit gates and checks.- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill includes a comprehensive 'never-store' policy in rules/privacy-and-consent.md that explicitly forbids the recording of sensitive data such as API keys, passwords, JWTs, private keys, and government IDs. It uses regex-based detection heuristics as a first line of defense during memory extraction.- [COMMAND_EXECUTION]: The skill describes using standard system utilities like grep, rm, and mv for managing memory files. These actions are scoped to specific, user-confirmed directory tiers (home, project-local, project-shared).- [PROMPT_INJECTION]: No malicious prompt injection patterns were detected. The skill uses instructional language to guide the agent in managing memory correctly rather than attempting to bypass safety filters.- [INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled markdown files, which is an inherent part of its functionality. This attack surface is mitigated by the mandatory 'Consent Preview' in the write pipeline, requiring the agent to surface proposed changes for user approval before they are persisted.

persistent-memory