IronClaw builds an operator inspector for live runs

nearai/ironclaw PR #7235 is the inspected primary source: “feat(inspector): add operator inspection API and live updates.” The related records below were inspected as supporting context rather than independent confirmation.

The facts

- nearai/ironclaw PR #7235 was created 2026-08-05T11:56:42Z and was open at inspection; its title is “feat(inspector): add operator inspection API and live updates.” - nearai/ironclaw PR #7277 was created 2026-08-06T10:54:02Z and was open at inspection; its title is “feat(inspector): add model call statistics.” - nearai/ironclaw PR #7291 was created 2026-08-06T14:03:53Z and was open at inspection; its title is “feat(inspector): complete statistics, navigation, and localization.” - The repository reported 12594 stars, 1516 open issues, default branch main, and last push 2026-08-07T00:53:18Z. - The latest tagged-release baseline checked was ironclaw-v1.1.0 published 2026-08-06T08:39:11Z.

What changed

nearai/ironclaw PR #7235 reports: Summary Add operator-only endpoints for retrieving bounded run snapshots, prompt diagnostics, and individual tool activity details. Add a live diagnostics stream with cursor-based resume, duplicate protection, retention rebasing, keepalive handling, and bounded connection lifetime. Preserve authenticated tenant and user scope through the transport and product query layers. Require both deployment-level operator support and caller-level operator permission before any diagnostic data is dispatched. Register the new routes through the existing descriptor and operator-route policy systems. Change Type [ ] Bug fix [x] New feature [ ] Refactor [ ] Documentation [ ] CI/Infrastructure [x] Security [ ] Dependencies Linked Issue Closes #7220 Depends on #7219 / #7230. Validation [x] cargo fmt --all -- --check [x] Targeted clippy checks completed with warnings denied [ ] cargo build: Not run separately because the affected packages were compiled by clippy and test commands [x] Relevant tests pass: Product contract

nearai/ironclaw PR #7277 reports: Summary Adds run-scoped model-call statistics to the Inspector, including call counts, latency, input/output token usage, cache token usage, and per-model breakdowns. Captures the effective provider model for each individual call so concurrent requests, fallbacks, and failed calls are attributed correctly. Clearly reports unavailable or partial usage evidence instead of treating missing metrics as zero. Keeps tool execution metrics out of the UI until tool-level diagnostic capture is implemented. Adds Rust, frontend, and browser E2E coverage. Browser E2E changes are isolated in a separate commit. Linked Issue Closes #7223 Depends on #7222 Part of #7218 Security Impact Diagnostic statistics remain operator-only, caller-scoped, bounded, and process-local. No credentials or unredacted prompt content are added to the statistics response. Database Impact None. No schema or migration changes. Blast Radius Limited to model-call diagnostic capture, the operator inspection API, and the Inspector statistics tab. Rollback Plan Revert this PR

nearai/ironclaw PR #7291 reports: Summary Adds total, successful, and failed tool-call statistics to the Inspector Stats tab. Adds direct latest-turn navigation and browser-observed stream health metrics with bounded, duplicate-safe session storage. Localizes Inspector labels, states, errors, metrics, buttons, and accessibility text, with complete English, Simplified Chinese, and Korean coverage. Updates the Inspector contract, user documentation, frontend tests, and browser E2E coverage. Linked Issue Closes #7287 Depends on #7226 Part of #7218 Validation [x] git diff --check [x] Frontend test suite: 1126 tests passed [x] Inspector focused suite: 21 tests passed [x] TypeScript typecheck and source convention checks [x] Vite production build [x] WebUI locale consistency tests: 5 passed [x] Inspector browser E2E scenarios [x] Manual testing of Prompt, Activity, Stats, navigation, stream metrics, and responsive presentation The existing bundle-budget check remains above its configured threshold at 218.7 KB on both this branch and the unchanged dependency branch. The Inspector

Why it matters

Agent runtimes need a way to answer what is running, which model calls dominate cost and where failures sit without granting a debug panel unlimited access. The open proposals still need authorization and payload-redaction scrutiny before becoming a trusted production cockpit. The operator test is whether authority, persisted state and the visible outcome still describe the same event after retries, restarts or delegation.

Current

The primary record was open when captured on 2026-08-07. The release baseline was ironclaw-v1.1.0 published 2026-08-06T08:39:11Z. Repository metadata, pull requests, issues, release records, Google News discovery, Hacker News discovery, Lobsters, Metamesh and the rendered ClawCharts row were checked. ClawCharts selected the subject; it did not prove the claim.

Evidence

The primary URL is https://github.com/nearai/ironclaw/pull/7235. Supporting records are https://github.com/nearai/ironclaw/pull/7277 and https://github.com/nearai/ironclaw/pull/7291. Source bodies, timestamps and states are preserved in the daily evidence bundle. Test counts, reproductions and deployment observations remain attributed to their authors unless explicitly identified as independently rerun.

Source boundary

Open work is described as open, closed work as closed, and operator reports as reports. A pull request is evidence of proposed or reviewed direction, not proof of a shipped release. Search residue, package mirrors and historically published source spines were excluded.

Operator take

Keep inspector access authenticated and read-only by default, bound event retention and response size, and verify that model statistics cannot expose prompts or credentials. Preserve a before-state receipt, make the smallest reversible change, and verify the original failure independently after intervention.

Caveat

Public project records are mutable. Status, scope and evidence can change after publication. This brief records the inspected state and does not authorize changes to a reader’s deployment.

Source inspected; source state, environment and release boundary remain explicit.