The Agent Skills Directory

[SAFE]: The skill implements a robust, read-only workflow for knowledge management. It prioritizes data integrity and confidentiality by requiring a private configuration file for routing facts to appropriate repositories based on confidentiality tiers.
[COMMAND_EXECUTION]: The skill uses a Python script scripts/kb_scan.py for repository reconnaissance. The script is implemented using only the Python standard library and operates in a read-only mode to scan markdown files for entity mentions. There are no patterns of arbitrary command execution or shell injection.
[DATA_EXFILTRATION]: While the skill manages knowledge bases, it does not perform network operations. It explicitly instructs the agent to stop if confidentiality routing configuration is missing and enforces that private facts should not be routed to public repositories.
[REMOTE_CODE_EXECUTION]: There are no remote code execution patterns. All operations are local to the configured knowledge base repositories, and dependencies are internal to the synthesis toolset.
[PROMPT_INJECTION]: The instructions focus on disciplined workflows and data reconciliation. There are no attempts to bypass safety filters or override system-level instructions.
[INDIRECT_PROMPT_INJECTION]: The skill processes session data, which is a potential surface for indirect injection. However, it mitigates this by requiring atomic claim extraction, mandatory pre-write scans, and human-in-the-loop validation checkpoints before shipping changes.

synthesis-knowledge-capture