The Agent Skills Directory

[SAFE]: The skill implements a robust security model centered on a local-first memory management approach. Analysis found no evidence of malicious obfuscation, hidden backdoors, or credential harvesting patterns.- [DATA_EXFILTRATION]: The skill's 'trust-boundary.md' explicitly prohibits network calls, webhooks, or telemetry during operation. It enforces a write allowlist limited to project memory files and repository-specific configuration (e.g., AGENTS.md, .cursor/rules), preventing arbitrary data leakage to external systems.- [CREDENTIALS_UNSAFE]: Guidelines within the skill strictly forbid the storage of secrets, tokens, private keys, or passwords in the memory files. Preflight hygiene scans are designed to maintain the integrity of these files.- [COMMAND_EXECUTION]: Shell operations are restricted to standard local search tools like 'ripgrep' (rg) and 'Select-String'. The skill explicitly prohibits executing remote scripts (e.g., curl | bash) during its execution pipeline.- [PROMPT_INJECTION]: The skill uses procedural instructions for memory management but does not include any patterns designed to override agent safety filters or bypass system instructions.- [INDIRECT_PROMPT_INJECTION]: While the skill ingests untrusted session data, it mitigates risk through a 'Consolidation protocol' that digests context, applies a quality filter (requiring facts to be concrete and verified), and uses a 'Conflict gate' to identify and resolve contradictions before writing to disk. Findings are paraphrased rather than directly copied to prevent imperative injection.

memo-session-skill