fix: telemetry improvements from deep AppInsights analysis

[anandgupta42]

What does this PR do?

Fixes telemetry gaps identified through deep analysis of altimate-code-os Azure AppInsights data (10-day window, 3,678 events, 8 machines):

1. Error classification — adds 4 new error classes (file_not_found, edit_mismatch, not_configured, resource_exhausted) to reduce "unknown" from 85%+ to ~50%
2. Session metadata — adds os, arch, node_version to session_start event for environment segmentation
3. Doom loop protection — adds per-tool call counter (threshold=30) to catch varied-input loops like todowrite 2,080x
4. Token visibility — adds tokens_input_total field for Anthropic where tokens_input excludes cached tokens
5. Query documentation — adds KQL reference documenting customDimensions vs customMeasurements

Type of change

• Bug fix (non-breaking change which fixes an issue)

Issue for this PR

Closes #586

How did you verify your code works?

• 189 telemetry tests pass (0 failures, 620 assertions)
• TypeScript typecheck passes (5/5 packages via turbo)
• Upstream marker guard passes
• Multi-model code review (5 models: Claude, Gemini 3.1 Pro, Kimi K2.5, MiniMax M2.5, GLM-5) — all actionable findings addressed
• Verified token data in live AppInsights via az monitor app-insights query

Checklist

• My code follows the project's coding standards
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally
• I have reviewed my own code
• I have run the linters and fixed any issues

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features

  • Added detection for repetitive tool execution with user alerts when repeat threshold is reached
  • Enhanced token usage tracking to include cached input tokens in total counts

• Improvements

  • Expanded system environment telemetry collection with OS, architecture, and runtime information
  • Refined error classification for better categorization

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review to trigger a review and subscribe this PR to future pushes, or @claude review once for a one-time review.

_{Tip: disable this comment in your organization's Code Review settings.}

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR extends telemetry event schemas with environment metadata (OS, architecture, Node version), improves error classification with four new error categories, introduces per-tool call counting to detect doom loops, and adds total input token accounting that includes cached tokens.

Changes

Cohort / File(s)	Summary
Telemetry Schema & Error Classification packages/opencode/src/altimate/telemetry/index.ts	Extended Telemetry.Event schema: added os, arch, node_version to session_start; added optional tokens_input_total to generation events; expanded core_failure error_class union with file_not_found, edit_mismatch, not_configured, resource_exhausted. Rewrote ERROR_PATTERNS to split error classifications and add dedicated keyword groups for each new error class. Added KQL documentation comment block.
Token Tracking packages/opencode/src/session/index.ts	Added inputTotal field to Session.getUsage() return value, computed as sum of adjusted input tokens plus cache read/write input tokens.
Doom Loop Detection packages/opencode/src/session/processor.ts	Introduced per-tool repeat call tracking via toolCallCounts map. When a tool's call count reaches threshold (30), emits doom_loop_detected telemetry event, issues PermissionNext.ask with permission: "doom_loop", and resets counter. Extended generation telemetry to conditionally include tokens_input_total when different from tokens_input.
Session Initialization packages/opencode/src/session/prompt.ts	Added environment metadata (os: process.platform, arch: process.arch, node_version: process.version) to session_start telemetry event in SessionPrompt.loop at step 1.
Test Updates packages/opencode/test/altimate/telemetry-signals.test.ts, packages/opencode/test/telemetry/telemetry.test.ts	Updated test payloads to include os, arch, node_version in session creation and event tracking. Reorganized error classification tests: moved patterns from "connection" group into dedicated not_configured, file_not_found, edit_mismatch, resource_exhausted test cases. Removed "moat" terminology from test descriptions.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~22 minutes

Possibly related PRs

• #445 — Extends Telemetry.Event schema with new event variants and fields in the telemetry subsystem.
• #336 — Modifies packages/opencode/src/session/processor.ts to adjust token accounting logic and per-generation telemetry.
• #245 — Updates core_failure error classification patterns and error_class union in telemetry schema.

Suggested labels

contributor

Suggested reviewers

• mdesmet
• kulvirgit

Poem

🐰 Telemetry now tracks where we hop and bound,
With OS, arch, and Node all neatly bound,
Doom loops detected before they spiral deep,
And token counts tallied—cached tokens don't sleep! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	Title clearly identifies the main change (telemetry improvements) and references the analysis source (AppInsights), directly aligning with the primary purpose of the PR.
Linked Issues check	✅ Passed	All coding objectives from #586 are met: error classification extended with 4 new classes (#586), session metadata added (os/arch/node_version #586), doom loop protection implemented with counter/threshold (#586), tokens_input_total field added (#586), and moat terminology removed (#586).
Out of Scope Changes check	✅ Passed	All changes directly support the linked issue objectives: telemetry schema/classification updates, session metadata collection, doom loop detection, token tracking improvements, and test file renaming—all within scope of fixing identified telemetry gaps.
Description check	✅ Passed	The PR description comprehensively covers the required template sections with detailed explanations of changes, verification methods, and complete checklist.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/telemetry-analysis-improvements

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}