AgentOS provider-aware routing

Deterministic compatibility, separate from control eligibility.

Modeled fit uses server-owned capabilities. It is not provider performance, assignment, approval, verification, dispatch, or execution.

Inspect routing analysis

Deterministic recommendationNot assigned

Claude Code

secure payment webhook

Highest modeled fit: 95/100
Recommendation state: deterministic recommendation
Recommendation use: available for human review
Routing analysis: routing-analysis-secure-payment-webhook
Task / repository: controlled-task-secure-payment-webhook / repository-connection-preview-secure-payment-webhook
Provider contacted: false
Dispatch / execution: disabled / disabled
Repository contents: not fetched
Evaluated: 2026-07-31T00:00:00.000Z

Recommendation use blockedNot assigned

Claude Code

harden risky prompt

Highest modeled fit: 93/100
Recommendation state: recommendation blocked
Recommendation use: blocked by revocation
Routing analysis: routing-analysis-harden-risky-prompt
Task / repository: controlled-task-harden-risky-prompt / repository-connection-preview-harden-risky-prompt
Provider contacted: false
Dispatch / execution: disabled / disabled
Repository contents: not fetched
Evaluated: 2026-07-31T00:00:00.000Z

Recommendation use blockedNot assigned

Cursor

investigate failing tests

Highest modeled fit: 95/100
Recommendation state: recommendation blocked
Recommendation use: blocked by kill switch
Routing analysis: routing-analysis-investigate-failing-tests
Task / repository: controlled-task-investigate-failing-tests / repository-connection-preview-investigate-failing-tests
Provider contacted: false
Dispatch / execution: disabled / disabled
Repository contents: not fetched
Evaluated: 2026-07-31T00:00:00.000Z

Operational review workspace · AI Work Ledger

Review AI software work with evidence, policy context, human approval, and a traceable record.

Review Runs bring intake, evidence, policy, risk, verification, approval, and merge recommendations into one controlled inspection surface.

Enterprise inspection surface: these Review Runs are the first AI Work Ledger v2 records. Review status shows where a review is in the safe processing lifecycle.

Review and recommendation only. Demo data is simulated; stored summaries remain read-only. No live execution, repo access, GitHub posting, automatic merge, or production persistence is enabled, and verification may be draft or simulated.

Open demo workspace Linked Tasks Applied Policies Verification Center Output Provenance Approval Center Governed Agent Context Workspace Attention Open Workspace Open MergeGuard View in Ledger Linked Work Record Explore Control Flow Review Analytics Open dashboard Beta guide Pilot demo First repo setup

Private beta workspace

Control before action

Human approval remains the decision boundary. Check and comment output are previews, and future GitHub actions remain not live in this product surface.

Read-onlySafe summarySimulated demo data

Review control lifecycle

From intake to traceable record

Each stage preserves the difference between observed evidence, system recommendation, draft decision, and human approval.

Review intake

Receive a safe work summary and bounded change context.

Collect evidence

Assemble report, test, policy, and preview signals.

Evaluate policy

Apply RepoBrain controls and required review flags.

Classify risk

Derive risk and lifecycle status deterministically.

Verify

Mark evidence passed, draft, incomplete, or needing review.

Human approval

Require an explicit human decision before merge.

Recommend

Produce a non-executing merge recommendation.

Record

Link the safe result to Work Record and Ledger foundations.

Derived workspace summary

Current review posture

Metrics are derived from the currently available safe summaries or deterministic demo records.

preview metrics

3Total runsDerived preview metric

0Awaiting reviewDerived preview metric

0Evaluator pending requirementsDerived preview metric

1VerifiedDerived preview metric

2Needs changesDerived preview metric

1BlockedDerived preview metric

17Evidence signalsDerived preview metric

Operational capabilities

Evidence and control context in one workspace

Review AI software work with evidence, policy context, human approval, and a traceable record.

Accept safe review intake from GitHub webhooks, imported diffs, and deterministic demos.

Collect report, repository-policy, check-preview, comment-preview, and warning evidence.

Evaluate RepoBrain policy context and protected or sensitive path matches.

Classify control risk without treating a recommendation as human approval.

Track verification state, required follow-ups, and human approval gates.

Produce a merge recommendation while automatic merge remains disabled.

Connect each review to a draft Agent Work Record and derived Ledger entry.

Keep future GitHub actions preview-only and disabled by default.

Review presets

Five controlled review starting points

Preset metadata is deterministic and non-persistent. Use demo mode to inspect representative Review Run records.

Load demo records

MergeGuardHIGH

Payment webhook security review

Inspect payment, session, and prompt changes with mandatory security approval.

MergeGuardMEDIUM

Risky dependency update

Review a major dependency bump, lockfile changes, tests, and rollback evidence.

Agent Work RecordsMEDIUM

UI refactor review

Check customer-visible behavior, component scope, accessibility, and test evidence.

AgentOSMEDIUM

Failing-test investigation

Separate observed failures from proposed fixes and verification evidence.

PromptForgeHIGH

Prompt-security change

Review injection, leakage, tool boundaries, approval gates, and security tests.

Simulated demo data

Demo mode uses simulated review runs. No real repo, customer data, GitHub posting, persistence, or live agents are used.

Review run summaries are unavailable.

acme-ai-checkout/acme-checkout-web

PR #42: Add AI refund assistant prompt and checkout webhook changes

Review status: Needs manual reviewHIGHneeds human reviewSimulated demo data

acme-ai-checkout/acme-checkout-web PR #42 is needs manual review with HIGH risk. Evidence is needs review. Human approval is required before merge. Source context: MergeGuard, PromptForge, AgentOS, Agent Work Records. PR #42 in acme-ai-checkout/acme-checkout-web is needs manual review with high control risk. Complete human review before merge.

Source: MergeGuard + PromptForge + AgentOS + Agent Work RecordsVerification: Needs reviewCurrent authority: Awaiting human approval

AI work summary

Complete human review before merge

MergeGuard report generated; RepoBrain rules fetched; 6 repo rules summarized; 5 policy checks triggered; 2 protected / 2 sensitive path matches; GitHub Check Preview: failure; PR Comment Preview v2 available; 3 safe warning labels

Human review required: yes
Risk level: High control risk
Merge recommendation: Needs human review

Review status: Needs manual review
Created: Jul 9, 09:00 AM
Mode: openai
Action: opened
Sender: demo-reviewer
Origin: github pr webhook
Work type: feature build
Scope: sensitive paths touched
Agent: Codex
Policy checks: 5
Violations: 1
Required follow-ups: 3
Recommended: 2
Repo rules: fetched
RepoBrain: 6 rules
Policy matches: 2 protected / 2 sensitive
Report: generated
Comment: dry-run
Check conclusion: failure
PR Comment v2: available

RepoBrain policy definition: Refund assistant and checkout webhook changes touch payment, prompt, and auth/session surfaces. Security and prompt review are required before merge.

GitHub Check Preview

ForgeLayer preview: merge blocked

Preview only. ForgeLayer is not creating GitHub check runs yet.

Conclusion: failure
Would block merge: yes
Check-preview policy flag: required

PR Comment Preview v2

ForgeLayer Review

Deterministic preview artifact, not current approval authority. GitHub posting is disabled.

Safe to post: no
Risk: HIGH
Recommendation: needs human review

paymentsprompt_securityauthtestsreview: securityreview: promptreview: code-owner

3 warnings

View safe detail Open PR

ForgeLayer

Operational review workspace

Loading Review Runs

Fetching the latest safe review-run summaries. Raw diffs, raw repo rules, secrets, tokens, and webhook payloads are not loaded into this view.

Read-only

Safe summaries

Preparing the capped private beta ledger view.

ForgeLayer

AgentOS provider-aware routing

Deterministic compatibility, separate from control eligibility.

Modeled fit uses server-owned capabilities. It is not provider performance, assignment, approval, verification, dispatch, or execution.

Inspect routing analysis

Deterministic recommendationNot assigned

Claude Code

secure payment webhook

Highest modeled fit: 95/100
Recommendation state: deterministic recommendation
Recommendation use: available for human review
Routing analysis: routing-analysis-secure-payment-webhook
Task / repository: controlled-task-secure-payment-webhook / repository-connection-preview-secure-payment-webhook
Provider contacted: false
Dispatch / execution: disabled / disabled
Repository contents: not fetched
Evaluated: 2026-07-31T00:00:00.000Z

Recommendation use blockedNot assigned

Claude Code

harden risky prompt

Highest modeled fit: 93/100
Recommendation state: recommendation blocked
Recommendation use: blocked by revocation
Routing analysis: routing-analysis-harden-risky-prompt
Task / repository: controlled-task-harden-risky-prompt / repository-connection-preview-harden-risky-prompt
Provider contacted: false
Dispatch / execution: disabled / disabled
Repository contents: not fetched
Evaluated: 2026-07-31T00:00:00.000Z

Recommendation use blockedNot assigned

Cursor

investigate failing tests

Highest modeled fit: 95/100
Recommendation state: recommendation blocked
Recommendation use: blocked by kill switch
Routing analysis: routing-analysis-investigate-failing-tests
Task / repository: controlled-task-investigate-failing-tests / repository-connection-preview-investigate-failing-tests
Provider contacted: false
Dispatch / execution: disabled / disabled
Repository contents: not fetched
Evaluated: 2026-07-31T00:00:00.000Z