Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/bearlike/assistant/claudegit clone --depth 1 https://github.com/bearlike/AssistantWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00000 | $0.03384 |
| Opus 5 | $0.00000 | $0.01692 |
| Sonnet 5 | $0.00000 | $0.00677 |
| Haiku 4.5 | $0.00000 | $0.00338 |
Grade A, and why
CLAUDE scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 223 lines — stays where its author put it; the contents beside it link to each section on GitHub.
agents/ — admission, delegation, and the agent lifecycle
Scope: hypervisor.py · spawn_agent.py · agent_context.py ·
attestation.py · agent_registry.py · built_in_agents.py. A closed
subsystem: these modules reference each other's types and are meaningless apart.
Owns automaton A of the three (parent file → "Three automata, not one"): the agent lifecycle, from admission to an absorbing terminal.
The lifecycle vocabulary is SIX members, four of them terminal
AgentStatus (hypervisor.py) is submitted · running · completed · failed · cancelled · rejected, sourced from A2A v1.0; the last four are absorbing.
ACTIVE_STATUSES = {submitted, running} is the non-terminal set, and every
hypervisor-side site (cleanup, cancel_pending, collect_running) reads that
constant rather than comparing against one literal. A2A spells the active state
working; this codebase spells it running — do not assume the two are
interchangeable when reading A2A material against this code.
Do NOT add a seventh member for "accepted but not started". A2A has no
queued/pending, and SUBMITTED is specified as "task acknowledged and
accepted", which already IS that state. The vocabulary is mirrored in three
clients (console agentStatus.ts, CLI agent_transcript_hub.py, Aura
TranscriptReducer.kt), so a new member costs five surfaces and buys a synonym.
agentStatusAlignment.test.ts pins the console mirror and is meant to fail if you
change the vocabulary.
Three RESULT vocabularies sit beside the lifecycle one:
| Vocabulary | Members | Answers |
|---|---|---|
AgentStatus |
6 | where a live agent IS |
AgentResultStatus |
5 | what the child REPORTED |
SpawnOutcomeStatus |
6 | how one spawn ATTEMPT ended |
AgentResultStatus.partial is minted nowhere. It is the one member able to
express "advanced the work but did not finish it", and no producer sets it —
OrchestrationState.terminal_status() returns completed/failed/cancelled by
declared type. So a budget-spent child and a crashed child reach the parent under
the same word, and the difference survives only in the stop event's detail and
the attestation's done_reason. SpawnOutcomeStatus deliberately is NOT a
superset of AgentResultStatus: it drops that dead member and reports the terminal
it SETTLED (SettledStatus). Minting partial for the budget-exhausted subset, or
deleting it, is a change to AgentResultStatus AND three client mirrors, not to an
alias. tests/test_child_settlement_parity.py pins the containment, so widening
any of these vocabularies fails there rather than silently changing what a spawn
can report.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 223 lines · 0 tokens per session scan A 38c891facc17
CLAUDE is an agent published in the GitHub repository bearlike/Assistant (41 stars, last pushed 8d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 3,384 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
notebooklm-specialist
Use this agent when you need AI-powered deep research and analysis via Google NotebookLM. Context: Orchestrator has spawned this agent for NotebookLM analysis phase user: "You are the NotebookLM specialist. Research brief: [topic]. Notebook ID: abc123. Output: ./docs/research/..." assistant: "Reading my skills and…
VSCode Tour Expert
Expert agent for creating and maintaining VSCode CodeTour files with comprehensive schema support and best practices.
Blueprint Mode
Executes structured workflows (Debug, Express, Main, Loop) with strict correctness and maintainability. Enforces an improved tool usage policy, never assumes facts, prioritizes reproducible solutions, self-correction, and edge-case handling.
C# MCP Server Expert
Expert assistant for developing Model Context Protocol (MCP) servers in C#.
agent-tools
Agent tools let one chat agent dispatch another chat-capable sub-agent as part of its work — per-run, parent-supervised delegation is the flagship use case for facet-backed dynamic agents. The child is a real sub-agent with its own Durable Object storage, messages, tools, resumable stream, and drill-in URL. The parent…
human-in-the-loop
Human-in-the-loop (HITL) patterns allow agents to pause execution and wait for human approval, confirmation, or input before proceeding. This is essential for compliance, safety, and oversight in agentic systems.