The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted user data which is later retrieved and interpolated into system prompts for the agent.
Ingestion points: User messages and external documents are processed via sdk.memories.create as seen in SKILL.md and reference/ingestion.md.
Boundary markers: The prompt injection templates in reference/context-fetch.md use Markdown headers such as '## Facts' but lack robust delimiters or instructions to ignore embedded commands.
Capability inventory: The integration involves network communication with Synap Cloud and the execution of subprocesses by the Node.js SDK.
Sanitization: Documentation in reference/context-fetch.md suggests filtering by a confidence score but does not specify content sanitization.
[DYNAMIC_EXECUTION]: The Node.js version of the SDK uses a bridge to execute Python code at runtime.
Evidence: reference/sdk-setup.md notes that the JS SDK spawns the Python SDK as a subprocess.
[COMMAND_EXECUTION]: The skill provides instructions for the user to execute shell commands to install dependencies and run verification scripts.
Evidence: SKILL.md and scripts/verify_synap.py involve package management commands and diagnostic script execution.
[DATA_EXFILTRATION]: The skill handles sensitive API keys and sends conversation data to external servers.
Evidence: The SYNAP_API_KEY is required for operations, and the SDK communicates with synap.maximem.ai to store and retrieve memory, as detailed in SKILL.md and reference/sdk-setup.md.

synap-codex