The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes raw content from Cognee memory documents to determine which records should be deleted. This content represents an untrusted data source that could contain malicious instructions designed to influence the agent's behavior during the evaluation process.
Ingestion points: Raw content fetched via ${CLAUDE_PLUGIN_ROOT}/scripts/cognee-forget.sh raw <dataset_id> <data_id> in SKILL.md.
Boundary markers: None identified. The agent is instructed to "Judge by meaning" without explicit delimiters or warnings to ignore embedded instructions in the raw data.
Capability inventory: The agent can perform deletions via cognee-forget.sh forget and execute arbitrary API calls using curl as described in the broader deletions section of SKILL.md.
Sanitization: No specific sanitization or filtering of the raw document content is performed before the agent processes it.
[DYNAMIC_EXECUTION]: The skill instructions utilize eval to load environment variables from the output of a local helper script.
Evidence: eval "$(${CLAUDE_PLUGIN_ROOT}/scripts/cognee-forget.sh env)" in SKILL.md. This pattern executes the output of the script in the current shell context to set configuration like COGNEE_BASE_URL and COGNEE_API_KEY.
[COMMAND_EXECUTION]: The skill performs shell command execution and network operations to interact with the Cognee server.
Evidence: Uses ${CLAUDE_PLUGIN_ROOT}/scripts/cognee-forget.sh for various operations (sync, datasets, data, raw, forget) and provides instructions for manual curl requests to the Cognee API endpoints.

cognee-forget