Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
human-in-the-loop-design - Skill - OpenSmartRoute
Skillv1.0.0
human-in-the-loop-design
See `content/06-decision-tree.xml`. The tree maps observable signals from inputs and intermediate artefacts to a rule from `01-core-rules.xml`, telling the agent which variant of the methodology to ap
Imported from faionfaion/faion-network (skills/faion/knowledge/ai-core/human-in-the-loop-design/AGENTS.md). Install upstream with npx skills add faionfaion/faion-network --skill human-in-the-loop-design. Copyright stays with the author.
Human-in-the-Loop Design
Summary
One-sentence: Spec for which AI decisions go to a human reviewer, how decisions are presented, how reviewer signals feed back into the model, and where the system blocks vs. proceeds.
One-paragraph: Spec for which AI decisions go to a human reviewer, how decisions are presented, how reviewer signals feed back into the model, and where the system blocks vs. proceeds. This methodology codifies the rules, output contract, failure modes, and decision tree needed for a spec produced by an agent applying human-in-the-loop design. The deliverable is validated against an explicit JSON Schema and routed through a decision tree that maps observable signals to rule ids in 01-core-rules.xml.
Ефективно для:
Building a reproducible spec for human-in-the-loop design across teams.
Reviewing AI-or-human work against an explicit contract instead of vibes.
Wiring the output into downstream automation (CI gates, observability, post-mortems).
Avoiding the failure modes listed in 03-failure-modes.xml.
Applies If (ALL must hold)
AI system makes decisions whose cost-of-error justifies a human review step (refund, content moderation, medical triage)
humans review at meaningful volume (not just escalations) and the review UX must be designed
reviewer feedback can flow back into model/prompt updates or routing rules
Skip If (ANY kills it)
AI decisions are low-stakes and rollback-cheap (autocomplete, ranking) — HITL is overhead
review volume is zero or near-zero (only escalations) — design the escalation path instead
review is purely audit (no decision changed by review) — use audit logging methodology
≥5 testable rules grounding the methodology with rationale + source
1100
content/02-output-contract.xml
essential
JSON Schema for the deliverable + valid/invalid/forbidden examples
900
content/03-failure-modes.xml
essential
≥3 antipatterns with symptom + root-cause + fix triplets
800
content/04-procedure.xml
essential
Step-by-step procedure end-to-end
800
content/05-examples.xml
essential
Worked example from real engagement
700
content/06-decision-tree.xml
essential
Routing tree → rule from 01-core-rules.xml
600
Task Routing
Sub-task
Model
Rationale
decision_inventory_map
sonnet
Map decisions to stakes + review need.
review_ui_spec
opus
Reviewer-facing UI spec + ergonomics.
feedback_loop_design
opus
How reviewer signals retrain / refit / route.
Templates
File
Purpose
templates/hitl-spec.md.j2
HITL spec skeleton
templates/hitl-spec.md
HITL spec skeleton Generated from templates/hitl-spec.md.j2 by tpl-jinja --migrate; do not hand-edit.
templates/review-decision.json
Review decision JSON schema
templates/_smoke-test.md.j2
Minimum viable filled-in HITL spec
templates/_smoke-test.md
Minimum viable filled-in HITL spec Generated from templates/_smoke-test.md.j2 by tpl-jinja --migrate; do not hand-edit.
Scripts
File
Purpose
When to call
scripts/validate-human-in-the-loop-design.py
Validate the spec artefact against the 02-output-contract schema
After subagent returns, before commit/publish
Related
[[ai-feature-progressive-rollout]]
[[ai-feature-incident-runbook]]
[[eval-driven-development-tdd-for-ai]]
Decision tree
See content/06-decision-tree.xml. The tree maps observable signals from inputs and intermediate artefacts to a rule from 01-core-rules.xml, telling the agent which variant of the methodology to apply or when to stop. Walk it on every fresh invocation; do not memo-ise outcomes across distinct engagements.
Use it
Copy one of these into your project. Installing also returns the manifest and these snippets.
yaml
targets:
- https://api.opensmartroute.ai/api/v1/registry/faionfaion-faion-network-human-in-the-loop-design/manifest # or paste the manifest below
Manifest
An Open Capability Manifest: the router reads it to know what this does, what it costs and when to pick it.
{
"ocm": "1",
"id": "faionfaion-faion-network-human-in-the-loop-design",
"kind": "skill",
"name": "human-in-the-loop-design",
"description": "See `content/06-decision-tree.xml`. The tree maps observable signals from inputs and intermediate artefacts to a rule from `01-core-rules.xml`, telling the agent which variant of the methodology to apply or when to stop. Walk it on every fresh invocation; do not memo-ise outcomes across distinct engagements.",
"publisher": "faionfaion",
"version": "1.0.0",
"capabilities": {
"domains": [
"general"
],
"tags": [
"skill-md",
"github-agents"
],
"languages": [
"en"
]
},
"quality_prior": 0.6,
"examples": [
"See `content/06-decision-tree.xml`. The tree maps observable signals from inputs and intermediate artefacts to a rule from `01-core-rules.xml`, telling the agent which variant of the methodology to apply or when to stop. Walk it on every fresh invocation; do not memo-ise outcomes across distinct engagements."
],
"primary": false,
"metadata": {
"source": {
"provider": "github-agents",
"repository": "https://github.com/faionfaion/faion-network",
"path": "skills/faion/knowledge/ai-core/human-in-the-loop-design/AGENTS.md",
"ref": "9f20bf6815773b771fae542acb362f864dbe49be",
"url": "https://github.com/faionfaion/faion-network/blob/9f20bf6815773b771fae542acb362f864dbe49be/skills/faion/knowledge/ai-core/human-in-the-loop-design/AGENTS.md",
"key": "faionfaion/faion-network/skills/faion/knowledge/ai-core/human-in-the-loop-design/AGENTS.md"
}
},
"instructions": "# Human-in-the-Loop Design\n\n## Summary\n\n**One-sentence:** Spec for which AI decisions go to a human reviewer, how decisions are presented, how reviewer signals feed back into the model, and where the system blocks vs. proceeds.\n\n**One-paragraph:** Spec for which AI decisions go to a human reviewer, how decisions are presented, how reviewer signals feed back into the model, and where the system blocks vs. proceeds. This methodology codifies the rules, output contract, failure modes, and decision tree needed for a spec produced by an agent applying human-in-the-loop design. The deliverable is va",
"cost": {
"context_tokens": 1161
}
}
Fetch it by URL: GET /api/v1/registry/faionfaion-faion-network-human-in-the-loop-design/manifest?version=1.0.0
Reviews
Star ratings from people who tried it. One review per account; edit yours any time.
No reviews yet. Install it, try it, and be the first to rate it.
