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/shrec/aiworkhub/agents-mdgit clone --depth 1 https://github.com/shrec/AIWorkHubWhat 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 | $0.01041 | $0.01041 |
| Opus 5 | $0.00521 | $0.00521 |
| Sonnet 5 | $0.00208 | $0.00208 |
| Haiku 4.5 | $0.00104 | $0.00104 |
Grade A, and why
AIWorkHub 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.
How it starts
The opening of the file, as written. The whole thing — 70 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AIWorkHub MCP Agent Notes
This is the standalone canonical AIWorkHub repository.
Rules:
- Keep every repository's task and context source of truth in that repository's
own
.aiworkhub/storage; never move task cards across repositories. - Use the native AIWorkHub MCP surfaces rather than legacy
AIToolsscripts or databases as model interfaces. - Write operations must remain explicit and gated by
AIWORKHUB_ALLOW_WRITES=1. - Never run
git add -Afrom this repository against the parent repository.
When printing worker auto-pickup or exact claim-start cards, use one self-contained fenced block per model. The instruction below must be the final non-empty line inside the same copyable block, never adjacent prose:
გაუშვი პარალელური სუბაგენტები და რომ დასრულდება დაუბრუნე კოდექსს რევიუსთვის
Target: AGENTS.md
AIWorkHub MCP tool-use policy
Order:
- validate the injected AIWorkHub Task MCP receipt, identity and scope.
- consume and acknowledge the injected project-context receipt.
- manager uses aiworkhub_manager_source_graph_query; worker uses aiworkhub_worker_source_graph_query.
- manager uses aiworkhub_manager_session_current_state; worker uses aiworkhub_worker_session_current_state.
- manager uses aiworkhub_manager_ai_memory_search; worker uses aiworkhub_worker_ai_memory_search.
- manager uses aiworkhub_manager_kb_search/get/related; worker uses aiworkhub_worker_kb_search/get/related.
- manager uses aiworkhub_manager_context_graph_search, aiworkhub_manager_context_graph_range and aiworkhub_manager_context_graph_related when enabled; workers never access Context Graph.
- execute exact card action and validation. Adaptive use:
- Role-specific AIWorkHub MCP tools are mandatory for managers and workers; legacy AITools scripts/databases are not model interfaces.
- Verified repo and repo_id outrank cwd, workspace_roots, environment_context and chat prose; on mismatch stop before filesystem access and switch/reload the route, never inspect the hinted repo as fallback.
- Task MCP receipt is always required; Source Graph is required for code tasks.
- Session Manager, AI Memory and KB run only when the card requests them or the task is non-trivial.
- Workers submit durable context changes only through the session/AI Memory/KB write-intent tools; a verified manager accepts or rejects each intent before canonical apply. Never write context databases directly.
- Do not make empty irrelevant calls to satisfy ceremony. Source Graph gate:
- When source_graph_required is true, stop if its bundle is unavailable, empty, stale or unacknowledged.
- Never use grep, rg, find, tree, broad cat/sed or recursive listing while Source Graph can index/process the target.
- A bounded exact-target fallback is allowed only after Source Graph reports that target unsupported or unindexed; record that reason.
- Re-query whenever the active symbol, dependency boundary, failure hypothesis, edit scope or validation target materially changes.
- Set workflow_stage on every Source Graph call: orientation, implementation, validation, review or rework; never relabel old calls after the fact.
- Start with focus/slice; escalate from returned evidence to context/calls/trace, impact, testmap/coverage and then a typed bundle only when needed.
- Use body for an exact symbol and bodygrep for indexed literal/body text; refresh once before any recorded bounded fallback.
- After Source Graph finds an exact target, prefer body/file preview; otherwise use a bounded read and never reread an unchanged range.
- For edits prefer aiworkhub_worker_semantic_edit_prepare/apply with the smallest verified range.
- Final HMAC-authenticated MCP audit ledger receipts distinguish injected, live, zero-hit and cache-hit calls plus modes and fallbacks; one preflight query is not continuous use. Exact-command exception:
- Exact validation/build/test commands named by the card are allowed.
- Exact known-path reads from the card or Source Graph are allowed; they are not broad discovery. Session Manager:
- Recover current state before non-trivial assumptions and preserve the returned session identity in the handoff.
- Never store secrets or fabricate session evidence. Manager Context Graph:
- Manager-only when enabled: search for non-trivial continuation, compaction/handoff recovery or prior-conversation facts; use range/related only from returned evidence.
- Workers never query or write Context Graph; durable context uses Session/AI Memory/KB write intents.
- Disabled or zero-hit is not failure; no empty ceremonial calls. AI Memory:
- After session recovery, issue one bounded task-specific query.
- Reuse returned durable decisions/lessons.
- Do not query legacy memory files directly. KB:
- Query authoritative project contracts/docs for unresolved factual context and preserve source identity.
- After a zero hit, do not repeat the query unless task scope changes. Stop at Codex review.
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 · 70 lines · 1,041 tokens per session scan A 7dbf57347304
AIWorkHub AGENTS.md is an instructions file published in the GitHub repository shrec/AIWorkHub (7 stars, last pushed 2d ago), licensed MIT. It adds 1,041 tokens to every session, about $0.0052 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
rondoflow CLAUDE.md
Instructions for rondoflow/rondoflow, covering rondoflow, project overview, tech stack, development and setup.
OpenVitamin AGENTS.md
Instructions for fengzhizi715/OpenVitamin, covering agents.md — inference gateway & plugin governance, 1. 项目与技术栈对齐说明, 2. 核心设计原则(core principles), 2.1 user-in-control(用户始终掌控) and 2.2 gateway-centric(以推理网关为中心).
hivemind CLAUDE.md
Instructions for dip497/hivemind, covering claude, research base (hyperresearch) — today is 2026-05-17, how to do research, what the skill files own and canonical research query.
hivemind AGENTS.md
Instructions for dip497/hivemind, covering agents.md — hivemind, what this is, monorepo layout, build / test / verify (run from apps/desktop unless noted) and conventions.
orcho-mcp AGENTS.md
Instructions for symphos-ai/orcho-mcp, covering orcho-mcp instructions, scope, local instruction files, workspace development pipeline and stable install is read-only.
forge CLAUDE.md
Instructions for SidCorp-co/forge, covering forge, workspace, commands, fourteen gates, six axes and doing the work.