The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's installation script fetches the library and its associated presets from the author's repository on GitHub.
Evidence: install.sh clones https://github.com/levi-qiao/longgraph-skill.git to a local cache directory.
[REMOTE_CODE_EXECUTION]: The documentation and installation script implement the pattern of piping a remote script to a shell for library setup.
Evidence: README.md recommends using curl -fsSL https://raw.githubusercontent.com/levi-qiao/longgraph-skill/main/install.sh | sh for installation.
[PERSISTENCE]: The skill uses host-specific features to maintain agent continuity across sessions, such as recurring tasks and background processes.
Evidence: grok.md utilizes scheduler_create for recurring tasks; codex.md uses automation actions for recurring wakeups; claude-code.md employs background sessions.
[COMMAND_EXECUTION]: The generated agent prompts are designed to execute various shell commands for testing, linting, and workspace management.
Evidence: The executor and supervisor templates include instructions for running tools like pytest, make lint, and git commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from workspace files, creating an attack surface where malicious data in those files could influence agent behavior.
Ingestion points: The agent nodes read from ledger.md, directives.md, and ops.md located in the user's workspace.
Boundary markers: The skill instructions include directives to disregard prior authoring context and follow a strict hierarchy of authority and 'red lines'.
Capability inventory: The agent is capable of shell command execution, background session management, and task scheduling.
Sanitization: Behavioral constraints and safety guidelines are provided within the prompts to mitigate risks from untrusted data.

longgraph