feat(mcp): evaluate when a rented loop should escalate to a human

[luciferlive112116]

Summary

Closes #4806

Implements the support/escalation-path decision logic: given an already-computed loop outcome, health tier, and operator/customer signals, decide whether a loop needs a human and what action to take — the deterministic core that routes "something's wrong" to a stop-and-review state. Composes with the loop-health evaluator (#4808) on the Rent-a-Loop path #4778.

• new packages/loopover-engine/src/loop-escalation.ts (pure): evaluateEscalation(input) returns shouldEscalate + action (none/notify/human_review/stop) + severity + reasons, by precedence — a requested stop wins; an errored run or critical health needs a human now; a give-up or customer flag needs a human soon; a soft degradation only notifies. No IO: it decides, the caller wires the action (a stop maps to Incident runbook for the kill-switch #4809's kill-switch once that lands).
• new loopover_evaluate_escalation MCP tool (src/mcp/server.ts); src/loop-escalation.ts is a thin re-export shim over the engine module.

Scope

• Conventional Commit title.
• Focused: the escalation evaluator + its tool, on the merged loop-lifecycle chain.
• Follows CONTRIBUTING.md; no site/CNAME/VitePress.
• Linked open issue (Closes Support & escalation path #4806, above).

Validation

• git diff --check
• npm run typecheck clean; engine build + build:mcp clean
• test:coverage on the changed code: loop-escalation.ts 100% lines & branches (24/24); the new src/mcp/server.ts handler + schema + registration fully covered (all 36 changed lines, both branch sides — diff-verified).
• MCP tool-invariant suites pass with the new tool (mcp-output-schemas "outputSchema on EVERY tool" + "schema-valid structured content").
• Tests cover every trigger and precedence tier plus the no-escalation path, engine-level and end-to-end through the MCP tool.

If any required check was skipped, explain why:

• Full test:ci not run end-to-end locally (Linux-only shell/self-host steps on Windows); the change-relevant gates above were validated directly.

Safety

• No secrets, wallets, hotkeys, trust scores, rewards, private rankings, or private maintainer evidence — the decision is public-safe loop status only, source-free.
• No auth/cookie/CORS/GitHub App/session change (pure function over caller-supplied data).
• MCP behavior added + tested (output schema + schema-valid content).
• No UI changes; no changelog edit.

Notes

• This is the deterministic decision half of Support & escalation path #4806; actually stopping a loop wires to Incident runbook for the kill-switch #4809's kill-switch once that lands (the stop action names it). The evaluator is independent of that wiring.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.11%. Comparing base (db74aa3) to head (b0e2c81).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5806   +/-   ##
=======================================
  Coverage   95.11%   95.11%           
=======================================
  Files         586      587    +1     
  Lines       46480    46509   +29     
  Branches    14860    14870   +10     
=======================================
+ Hits        44209    44238   +29     
  Misses       1515     1515           
  Partials      756      756           

Flag	Coverage Δ
shard-1	43.76% <86.20%> (+0.02%)	⬆️
shard-2	36.18% <55.17%> (-0.06%)	⬇️
shard-3	32.12% <10.34%> (+0.04%)	⬆️
shard-4	32.81% <10.34%> (-0.02%)	⬇️
shard-5	31.62% <10.34%> (-0.02%)	⬇️
shard-6	44.51% <10.34%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
packages/loopover-engine/src/loop-escalation.ts	100.00% <100.00%> (ø)
src/mcp/server.ts	95.94% <100.00%> (+0.03%)	⬆️

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[loopover-orb]

Tip

🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩

✅ LoopOver review result - approve/merge recommended

_{Review updated: 2026-07-14 19:11:11 UTC}

6 files · 1 AI reviewer · no blockers · readiness 100/100 · CI green · clean

✅ Suggested Action - Approve/Merge

• safe to merge

Review summary
This adds a small, pure escalation-decision function (`evaluateEscalation`) plus wiring through the engine barrel, a re-export shim, and a new MCP tool, mirroring the existing quota/loop-health evaluator pattern in this codebase. The precedence logic (kill > error/critical > abandoned/customer-flagged > degraded > none) is straightforward and correctly implemented as an if/else-if chain, and the accompanying tests exercise every branch and precedence tier at both the engine and MCP-tool level. The `severity`/`action` output schema fields are all `.optional()` in Zod despite always being populated by `evalEscalation`, which is a minor schema-looseness nit rather than a defect.

Nits — 5 non-blocking

• src/mcp/server.ts's `evaluateEscalationOutputSchema` marks `shouldEscalate`, `action`, `severity`, and `reasons` all `.optional()`, even though `evaluateEscalation` always returns every field — tightening this to required would give MCP clients a stronger contract, consistent with how confidently the tool always returns a complete decision.
• The external brief's 'deep nesting depth 5' flag on loop-escalation.ts:56 is just a linear if/else-if chain for 5-way precedence, not a real nesting concern — not actionable.
• `packages/loopover-engine/src/index.ts` and `src/mcp/server.ts` being flagged as long files is pre-existing accumulation, not something introduced by this diff's ~9 and ~36 line additions respectively.
• Consider making the four fields in `evaluateEscalationOutputSchema` (src/mcp/server.ts) required rather than optional, since `evalEscalation` never omits any of them.
• If a future PR wires the actual `stop` action to Incident runbook for the kill-switch #4809's kill-switch, make sure the MCP tool's 'it decides, caller wires' contract stays documented at the call site so the separation of concerns doesn't erode.

Signal	Result	Evidence
Code review	✅ No blockers	1 reviewer
Linked issue	✅ Linked	#4806
Related work	✅ No active overlap found	No same-issue or scoped active PR overlap found.
Change scope	✅ 20/20	Low review scope from cached public metadata (1 linked issue).
Validation posture	✅ 25/25	PR body includes validation/test evidence.
Contributor workload	✅ 10/10	Author activity: 95 registered-repo PR(s), 42 merged, 39 issue(s).
Contributor context	✅ Confirmed Gittensor contributor	luciferlive112116; Gittensor profile; 95 PR(s), 39 issue(s).
Gate result	✅ Passing	No configured blocker found.
Improvement	✅ Minor	risk: clean · value: minor — Code changes are accompanied by test evidence. LLM value judgment: moderate — The PR delivers a small, well-tested, pure decision function plus its MCP tool that closes the linked issue #4806 and slots cleanly into the existing loop-lifecycle evaluator pattern (quota/loop-health) without unrelated scope creep.

Linked issue satisfaction

Partially addressed
The PR builds solid deterministic decision logic for when/how a loop should escalate (a real building block for the escalation path) but it is explicitly blocked on #4809's kill-switch and does not implement or simulate an actual end-to-end escalation reaching a human, nor demonstrate the acceptance criterion of reaching a human-reviewable state within a defined target time.

Review context

• Author: luciferlive112116
• Role context: outside_contributor
• Public audience mode: oss maintainer
• Lane context: Repository is configured for direct PR review.
• Public profile languages: not available
• Official Gittensor activity: 95 PR(s), 39 issue(s).
• PR-specific overlap: none found.

Contributor next steps

• Keep the PR focused and include validation evidence before maintainer review.

Signal definitions

• Related work = same linked issue, overlapping active PRs, or title/path similarity.
• Change scope = cached public metadata such as size labels, draft state, and review-burden hints.
• Validation posture = whether the PR provides enough public validation/test evidence for maintainer review.
• Contributor workload = public contributor activity and cleanup pressure, not a repo-wide quality failure.
• Contributor context = public GitHub/Gittensor identity context; non-Gittensor status is not a blocker.

[BETA] Chat with Gittensory

Ask Gittensory a question about this PR directly in a comment — grounded only in the same cached, public-safe facts shown above, never a new claim.

• @gittensory ask <question> answers contribution-quality Q&A with source citations and freshness.
• @gittensory chat <question> answers in natural prose from cached decision-pack facts via local inference (maintainer/collaborator; read-only).
• A plain-language @gittensory mention with a real question is routed to the closest matching read-only command automatically -- no exact syntax required.

Full command reference: https://gittensory.aethereal.dev/docs/gittensory-commands

_{🟩 Safe / merged · 🟦 Advisory · 🟨 Held for review · 🟥 Blocked / closed}

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by LoopOver, a quiet PR intelligence layer for OSS maintainers.

• Re-run LoopOver review

[loopover-orb]

Gittensory approves — the gate is satisfied and CI is green.