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 instructions/ranjiang23/codex-hermes-worker/agents-mdgit clone --depth 1 https://github.com/RANJIANG23/codex-hermes-workerWhat 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 | $0.00369 | $0.00369 |
| Opus 5 | $0.00185 | $0.00185 |
| Sonnet 5 | $0.00074 | $0.00074 |
| Haiku 4.5 | $0.00037 | $0.00037 |
Grade A, and why
codex-hermes-worker AGENTS.md 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.
What it actually says
Codex + Hermes/Qwen routing rules
Delegate to the local worker first
- Batch file inspection, filename or directory classification, asset triage, string grouping.
- First-pass pseudocode summaries, candidate function names, metadata normalization.
- Repetitive evidence collection, result conversion, first-pass cross-checking, and low-value bulk summaries.
Use hermes_health before the first delegation. Prefer submit_local_job for batches and read get_local_job_summary before querying individual records.
Keep with GPT-5.6
- Overall reverse-engineering strategy, unknown compression/encryption/obfuscation, parser design and debugging.
- Cross-subsystem reasoning, decompiler correction, adjudicating conflicting hypotheses, final validation and reporting.
Escalate to GPT-5.6
Escalate when confidence is below 0.70, workers conflict, deterministic evidence conflicts with a model conclusion, an unknown encoding or file-format core field is involved, size/integrity checks fail, step limits are hit, schema validation fails, or needs_review=true.
Context control
Read aggregate summaries, conflicts, low-confidence items, and the review manifest first. Do not load an entire batch into the primary context by default.
Safety
Use restricted_batch for research/game/source data and treat those inputs as read-only. Its bridge tools may write only below this project's work directory.
delegate_trusted_full_task is a separate high-risk path. Use it only when the user explicitly authorizes broad host tools for the current task, pass authorization="explicit_user_authorized", and keep allow_network=false unless external access is also explicitly authorized. This mode can run an unsandboxed host terminal and modify files, so never select it merely because restricted execution is inconvenient.
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 · 29 lines · 369 tokens per session scan A 1153ee078e65
codex-hermes-worker AGENTS.md is an instructions file published in the GitHub repository RANJIANG23/codex-hermes-worker (0 stars, last pushed 1mo ago), licensed MIT. It adds 369 tokens to every session, about $0.0018 per session on Opus 5. 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 instructions, from other repositories
E2E Harness Self-Fixing Protocol
Rules for working with the Sir Thaddeus E2E test harness. Governs how to diagnose, fix, and verify failures without gaming the scoring system.
Task Execution Workflow
Rules for selecting, implementing, testing, and closing work in the Sir Thaddeus repository without requiring an external task tracker.
obsidian-llm-hub AGENTS.md
Instructions for takeshy/obsidian-llm-hub, a project described as: Obsidian plugin for AI chat, workflow automation, and semantic search — supports Gemini, OpenAI, OpenRouter, Grok, Ollama, and CLI backends.
offload-harness CLAUDE.md
Instructions for dmmdea/offload-harness, covering claude.md — agent orientation map for offload-harness, components & ports, model tiers (served by llama-swap on :11436), golden commands (all verified on this machine) and install / verify the stack (windows).
hermit-agent CLAUDE.md
Instructions for cafitac/hermit-agent, covering hermit project — mcp executor, key modules, public cli, tests and product boundary.
portable-agent-skills AGENTS.md
Instructions for ch1109/portable-agent-skills: Read CONTRIBUTING.md before changing a Skill.