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 commands/cafitac/hermit-agent/feature-develop-hermitgit clone --depth 1 https://github.com/cafitac/hermit-agentWhat 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 | $0.02354 |
| Opus 5 | $0.00000 | $0.01177 |
| Sonnet 5 | $0.00000 | $0.00471 |
| Haiku 4.5 | $0.00000 | $0.00235 |
Grade A, and why
feature-develop-hermit 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 — 260 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/feature-develop-hermit — Claude plans, Hermit implements
Same outline as /feature-develop, but the Phase 3 implementation
is delegated to Hermit over MCP so Claude's context never absorbs the
grunt-work tool output.
Claude is the orchestrator (interview, pattern search, plan, final verification). Hermit is the executor (TDD red-green implementation).
Arguments
$ARGUMENTS — PR number (or a short slug if you are not using GitHub PRs).
--strategy {aggressive|moderate|conservative}— delegation depth. If omitted, inferred from the model-strategy map in the checkpoint config (see "Checkpoint policy").--model <model_name>— which executor model Hermit should run with (e.g.--model glm-5.1,--model qwen3-coder:30b). If omitted, the gateway's default model is used.
Checkpoint policy
The delegation depth is read from:
$HERMIT_CHECKPOINT_CONFIG (env override)
├── ~/.claude/state/hermit-checkpoints.json (user-local)
└── scripts/harness/hermit-checkpoints.default.json (repo default)
Three strategies
| Strategy | Claude does | Hermit does |
|---|---|---|
aggressive |
interview · plan review · final verify | pattern search · plan doc · implementation |
moderate |
interview · pattern search · plan review · final verify | plan doc · implementation |
conservative |
interview · pattern search · plan doc · plan review · final verify | implementation only |
Auto-selection
If --strategy is absent, query the gateway's active model
(./bin/gateway.sh --status or GET /health → active_models[0]) and look
it up in the config's model_strategy_map:
- Prefix or substring match (e.g.
glm-5.1→glm→aggressive) - If nothing matches, fall back to
default_strategy.
Prerequisites
ghCLI authenticated (or skip the GitHub steps if you don't use it)- Hermit gateway + MCP server running (
./bin/gateway.sh --status) - Current branch is the PR branch, or the worktree already exists
Workflow
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 · 260 lines · 0 tokens per session scan A e8d449c28918
feature-develop-hermit is a command published in the GitHub repository cafitac/hermit-agent (4 stars, last pushed 6d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 2,354 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 2026-08-31.
Other commands, from other repositories
toefl-grade
TOEFL 답안/녹음 채점 — 로컬 LLM(Ollama)과 whisper.cpp로 4개 영역을 채점하여 SCORES.md에 점수를 누적 기록한다.
toefl-practice
TOEFL 유형 문제 생성 — 로컬 LLM(Ollama)이 reading/listening/speaking/writing 영역의 실전형 문제를 생성하여 practice/에 저장한다.
toefl-roadmap
이번 주 토플 학습 로드맵/과제 표시 — schedule.yaml에서 사용자가 지정한 학습 기간·요일·시간대·주차별 목표를 읽어 현재 날짜 기준으로 안내한다.
toefl-status
토플 학습 현황 요약 — schedule.yaml(D-day/목표점수) + JOURNAL.md/SCORES.md를 파싱하여 진행률, 영역별 점수 추이, 목표 대비 갭, 시험일까지 남은 일수를 표시한다.
toefl-drill
취약 영역 집중 반복 — SCORES.md에서 약점을 식별하여 해당 유형의 문제를 연속 생성/채점하는 딥 드릴 루프를 안내한다.
opencode
Manage OpenCode/Zen connections, switch models, and configure local Ollama.