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/scaffoldic/agentforge-py/agents-mdgit clone --depth 1 https://github.com/Scaffoldic/agentforge-pyWhat 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 | $0.02434 | $0.02434 |
| Opus 5 | $0.01217 | $0.01217 |
| Sonnet 5 | $0.00487 | $0.00487 |
| Haiku 4.5 | $0.00243 | $0.00243 |
Grade C, and why
agentforge-py AGENTS.md scanned grade C with 1 finding 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.
Hidden instructionshighPrompt injection
Directives inside HTML comments, invisible characters or bidirectional overrides are read by the model and not by the person reviewing the file.
<!-- Project-specific instructions go below this line. Survives upgrades. --> How it starts
The opening of the file, as written. The whole thing — 186 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — agentforge-py
Repository conventions for any AI assistant editing this repo. Tool-agnostic —
CLAUDE.mdand any future tool-specific rules files defer to this one.
What this repo is
agentforge-py is the Python implementation of AgentForge,
an open-source plug-and-play framework for building production AI
agents. The repo is a uv workspace with two member packages today:
packages/agentforge-core/— locked contracts (ABCs, value types, production-rails primitives, resolver, testing utilities). No I/O, no third-party SDKs except Pydantic + python-ulid.packages/agentforge/— default runtime (Agentorchestrator,InMemoryStore, configuration loader). Imports fromagentforge-core; never the reverse.
When new modules ship (provider clients, persistence drivers, MCP,
observability, safety, etc. — see CHANGELOG.md), each lands as a
new directory under packages/. The workspace glob in the root
pyproject.toml already covers them.
Hard rules
| # | Rule | Reference |
|---|---|---|
| 1 | agentforge-core imports nothing from agentforge or any other module package. It is the leaf of the dependency graph. |
dependency policy |
| 2 | Contracts in agentforge-core (every ABC + the Finding Protocol + the locked value types) are stable surface. Adding a method to an ABC is a major version bump; adding a field with a safe default is a minor bump. |
locked contract layer |
| 3 | Configuration is data, not code. No dynamic imports from YAML; no Jinja inside config. Env-var interpolation only (${VAR}, ${VAR:default}, ${VAR:?error}, $$ → $). |
agentforge.config |
| 4 | No magic numbers in production code. Every threshold / timeout / limit comes from a Pydantic config model with a documented default. | configuration policy |
| 5 | Test coverage must be ≥ 90% on every commit. Pre-commit blocks below; CI ratchet rejects regressions on main. |
.pre-commit-config.yaml + .github/workflows/ci-linux.yml |
| 6 | One feature = one branch = one PR. Conventional Commits format (feat:, fix:, docs:, test:, refactor:, chore:, perf:, revert:). |
PR template |
| 7 | Never bypass pre-commit with --no-verify unless the user explicitly authorises it for a specific commit (and the bypass is documented in the commit message). |
pre-commit policy |
| 8 | Type hints everywhere. mypy --strict is the gate. Any only at genuine boundaries (raw provider responses); never to paper over untyped internals. |
pyproject.toml > [tool.mypy] |
| 9 | Async-first. Public methods on locked contracts are async def. Sync callers use the *_sync shims (where exposed) or asyncio.run(). |
locked contract layer |
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 · 186 lines · 2,434 tokens per session scan C b5c99faf8880
agentforge-py AGENTS.md is an instructions file published in the GitHub repository Scaffoldic/agentforge-py (10 stars, last pushed 13d ago), licensed Apache-2.0. It adds 2,434 tokens to every session, about $0.0122 per session on Opus 5. A static security scan graded it C with 1 finding (hidden instructions). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
eigent AGENTS.md
AGENTS.md instructions for eigent-ai/eigent, covering eigent agent instructions, ui work, product terminology and validation.
LeAgent AGENTS.md
Instructions for vixues/LeAgent, covering leagent development guidelines, project overview, architecture, execution topology (one kernel, many ingresses) and surface map (where things live).
HARTOS CLAUDE.md
Instructions for hertz-ai/HARTOS, covering claude.md, project overview, master key - ai exclusion zone, branch discipline — main branch only (mandatory) and sibling repos — canonical filesystem paths.
semantix CLAUDE.md
Instructions for Gnosil/semantix, covering 项目规则（claude code） and 1. 改动须经用户过目；由 claude 提交到分支 + pr，用户 review 后合并.
openacme CLAUDE.md
Instructions for sandydasari/openacme, covering openacme — claude code guide, workspace layout, the agent loop — request path, why sse-only and custom data parts.
orloj AGENTS.md
Instructions for OrlojHQ/orloj, covering orloj agent instructions, must-follow sync rules and working style.