Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →lingtai-ai/tutorial-guide-agent-runtime
Nested tutorial-guide reference for lessons 4–6: init.json, lingtai-agent run, heartbeat and signal files, TUI runtime wrapping, and system prompt identity.
Passed 6 of 14 test cases with the skill, and 3 of 14 without it.
and 5 more in the full report
7 fixed3 passed either way11 still failing2 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
Write the exact Python one-liner or snippet needed to inspect SystemPromptManager's default section render order from the kernel module.
| Checked | Without skill | With skill |
|---|---|---|
| Imports SystemPromptManager from lingtai.kernel.prompt or lingtai_kernel.prompt. | Missed | Passed |
| Prints SystemPromptManager._DEFAULT_ORDER. | Missed | Passed |
The instructions your model reads when the skill loads.
Nested tutorial-guide reference for agent runtime lessons 4–6, reached from the root tutorial-guide router. Teach live per the router's discover, don't recite rule: read the real file or run the real command before explaining it.
lingtai-agent runRead YOUR init.json and walk through every field you find. Do not recite a list of fields — read the file and explain what is there. The human sees the real structure.
Key patterns to explain:
_file convention: live init fields like covenant, pad, comment, base_prompt take inline text or a <field>_file path to a shared file. (Older fields such as principle, procedures, brief, soul, and the legacy prompt are migrated by the kernel but no longer seeded here.) Note: there is no seed field for the agent's character in init.json — character is durable state the agent authors for itself after creation, managed via system/lingtai.md / psyche, not written into init.json.manifest contains: llm, agent_name, language, capabilities, admin, etc.addons connects to external messaging services.env_file for secrets.lingtai-agent runExplain the boot sequence: read init.json → load env → resolve venv → build Agent → clean stale signals → install signal handlers → start in ASLEEP state → agent.start() blocks on shutdown.
Emphasize: the agent is a long-running Python process. It does not exit after one task.
Show your own .agent.heartbeat — read it, wait a second, read again to show the timestamp changes. Explain the signal files: .interrupt, .suspend, .sleep, .prompt.
Explain: lingtai-tui is a Go frontend, not the agent. It creates agents (writes init.json), launches them (python -m lingtai run), monitors them (.agent.heartbeat, .agent.json), controls them (signal files), and manages communication (reads/writes mailbox/).
Draw the architecture diagram (TUI ↔ filesystem ↔ agent process).
Explain: you do not need the TUI to run an agent. A valid init.json + lingtai-agent run is sufficient.
Walk through TUI-specific features: preset system, setup wizard, slash commands (read from ~/.lingtai-tui/commands.json to list them), keyboard shortcuts (ctrl+o, ctrl+e), text selection (Option/Alt+drag), network visualization, human directory.
Mention that headless CLI management commands exist; Lesson 10 walks through lingtai-tui --help in full, so leave the enumeration for then.
test cases passed, vs 3 / 14 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.