fix(composio/triage): demote provider-config-rejection rollups (Sentry TAURI-RUST-1V)

[oxoxDev]

Summary

• Routes [composio][triage] run_triage failed through the central observability classifier instead of raw tracing::error! — so user-config / budget-exhausted rollups from the upstream provider chain get demoted to info-level breadcrumbs.
• Adds "may not be available on your provider" to is_provider_config_rejection_message — the canonical phrase emitted by reliable.rs:332 when the user's reliability.model_fallbacks chain is misconfigured.
• Net effect: drops the bulk of self-hosted Sentry TAURI-RUST-1V (10,692 events / 14d) where the inner provider attempts the provider layer already demoted were re-surfaced by the outer composio re-emit.

Problem

Self-hosted Sentry tauri-rust project's #4 unresolved by event count (14d, sort=freq) is [composio][triage] run_triage failed at 10,692 events. Breadcrumbs show every event boils down to:

[llm_provider] native_chat budget-exhausted 400 — not reporting to Sentry          (correct demote, inner)
Non-retryable error, moving on
Exhausted retries, trying next provider/model
llm_provider.reliable_chat skipped expected budget-exhausted error: The model `chat-v1` may not be available on your provider. Configure a fallback chain via `reliability.model_fallbacks` in …    (correct demote, inner)
[triage::evaluator] agent turn dispatch failed
[composio][triage] run_triage failed       ← raw tracing::error! emits a Sentry error event

The reliable-provider stack and the inference observability layer already classify these as user-config — but the bus-level re-emit at memory_sync/composio/bus.rs:354 used a raw tracing::error! that bypassed the classifier entirely.

Root cause is user-side: the user's reliability.model_fallbacks config doesn't list a model the provider actually serves. Remediation = fix that config. Sentry has no remediation path.

Solution

Two surgical changes, kept as separate micro-commits for review:

1. src/openhuman/inference/provider/config_rejection.rs — append "may not be available on your provider" to the PHRASES table feeding is_provider_config_rejection_message. Canonical phrase, anchored to reliable.rs:332 (the sole producer in-tree). Comment cross-links the call site so future drift of that wording is caught at review.

2. src/openhuman/memory_sync/composio/bus.rs — swap the raw tracing::error! at L354 with crate::core::observability::report_error_or_expected(..., "composio", "trigger_triage", &[("label", …)]). Same structured fields as before, but now the classifier runs. Genuine runtime bugs that don't classify still surface as full Sentry errors.

The two changes are additive; either landed alone would partially help, but only the pair fully closes the loop.

Submission Checklist

• Tests added or updated (happy path + at least one failure / edge case) per Testing Strategy
• Diff coverage ≥ 80% — changed lines (Vitest + cargo-llvm-cov merged via diff-cover) meet the gate enforced by .github/workflows/coverage.yml. Run pnpm test:coverage and pnpm test:rust locally; PRs below 80% on changed lines will not merge.
• N/A: behaviour-only change — classifier anchor + call-site swap, no new user-facing feature row. (Coverage matrix updated for added/removed/renamed feature rows.)
• All affected feature IDs from the matrix are listed in the PR description under ## Related
• No new external network dependencies introduced (mock backend used per Testing Strategy)
• N/A: observability classifier internals — no release-cut surface touched. (Manual smoke checklist not required.)
• N/A: Sentry-only fix — no GitHub issue. Sentry-Issue: TAURI-RUST-1V is in ## Related instead of Closes #NNN.

Impact

• Runtime/platform: desktop only. Demoted events still appear as info! breadcrumbs in local trace, so support can still inspect them. Sustained outages — were they ever to indicate a real triage bug — would surface via separate health/escalation paths.
• Performance: negligible (classifier already runs on the surrounding paths; one extra substring scan per failed triage turn).
• Security/migration: none.
• Compatibility: no public API change.

Related

• Sentry-Issue: TAURI-RUST-1V (self-hosted sentry.tinyhumans.ai → tauri-rust project, 10,692 events / 14d)
• Pattern precedent: fix(observability,composio): demote direct-mode Composio 401/Invalid API key noise (Sentry TAURI-RUST-X9) #2481 (composio direct-mode 401 demote), fix(observability,embeddings): demote Ollama embed user-config rejections (Sentry TAURI-RUST-XS + MA/KM/GX) #2612 (Ollama embed user-config demote), fix(app): guard CEF IPC fallback sync throw via safeInvoke wrapper (Sentry TAURI-REACT-7 + 6) #2619 (CEF IPC fallback guard)
• Closes:
• Follow-up PR(s)/TODOs:
  • Companion PRs for self-hosted Sentry TAURI-RUST-15 (Discord supervision restart-storm, 11.4k events) and TAURI-RUST-34H (Cloudflare anti-bot 500 wrap, 8.9k events) — coming behind this one to keep PR surface small.
  • Phase 7 Step 4.5 auto-resolve grep regex currently only matches OPENHUMAN-(TAURI|REACT|CORE)-… and BACKEND-ALPHAHUMAN-… shapes; self-hosted prefix is TAURI-RUST-…. Widen the regex (separate workflow-tooling PR) so the post-merge Sentry resolve sweep finds these IDs.

---

AI Authored PR Metadata (required for Codex/Linear PRs)

Linear Issue

• Key: N/A: Sentry-only — no Linear issue
• URL: N/A: Sentry-only — no Linear issue

Commit & Branch

• Branch: fix/sentry-1v-composio-triage-noise
• Commit SHA: $(git rev-parse HEAD)

Validation Run

• N/A: no app/ files touched — pnpm --filter openhuman-app format:check not needed.
• N/A: no app/ files touched — pnpm typecheck not needed.
• Focused tests: `cargo test --lib openhuman::inference::provider::config_rejection` (7/7 passed)
• Rust fmt/check (if changed): `cargo fmt --check` clean; `cargo check --manifest-path Cargo.toml` clean (pre-existing warnings only, not from this PR)
• N/A: no Tauri shell files touched — Tauri fmt/check not needed.

Validation Blocked

• `command:`
• `error:`
• `impact:`

Behavior Changes

• Intended behavior change: Composio-triage failures whose underlying err already classifies as user-state via the central observability classifier now demote to info-level breadcrumbs instead of emitting Sentry errors.
• User-visible effect: None — UI surfaces of the underlying conditions (budget-exhausted toasts, model-unavailable banners) are emitted at the inner layers and unchanged. Only the Sentry noise drops.

Parity Contract

• Legacy behavior preserved: All structured fields previously logged (`label`, `error`) are preserved in the routed report's message and tag set. Genuine runtime errors that don't classify still reach Sentry.
• Guard/fallback/dispatch parity checks: `is_provider_config_rejection_message` continues to require an anchored substring match — false-positive surface is identical to existing entries; new entry is the canonical `reliable.rs:332` remediation phrase.

Duplicate / Superseded PR Handling

• Duplicate PR(s): N/A: no duplicates
• Canonical PR: N/A
• Resolution (closed/superseded/updated): N/A

Summary by CodeRabbit

• Bug Fixes

  • Improved detection of provider-configuration rejection messages so fallback remediation notices are correctly classified as provider-config issues.

• Chores

  • Updated triage error handling to route failures into structured observability/reporting for clearer diagnostics and monitoring.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: b61f1e67-db9b-46de-bb93-0cb18c46a4f5

📥 Commits

Reviewing files that changed from the base of the PR and between e8397ab and 1733f06.

📒 Files selected for processing (2)

• src/openhuman/inference/provider/config_rejection.rs
• src/openhuman/memory_sync/composio/bus.rs

🚧 Files skipped from review as they are similar to previous changes (2)

• src/openhuman/inference/provider/config_rejection.rs
• src/openhuman/memory_sync/composio/bus.rs

---

📝 Walkthrough

Walkthrough

Adds "may not be available on your provider" to the provider config-rejection matcher with a unit test; routes Composio triage errors through crate::core::observability::report_error_or_expected instead of direct tracing::error!.

Changes

Error Classification and Observability

Layer / File(s)	Summary
Provider config-rejection detection src/openhuman/inference/provider/config_rejection.rs	Adds the substring "may not be available on your provider" to PHRASES and adds detects_reliable_chain_exhaustion_rollup test asserting both the multi-line rollup wrapper and inner remediation match is_provider_config_rejection_message.
Composio triage observability routing src/openhuman/memory_sync/composio/bus.rs	Replaces direct tracing::error! with crate::core::observability::report_error_or_expected in run_triage error handling, reporting a formatted detail string with category/context ("composio", "trigger_triage") and a label attribute.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• tinyhumansai/openhuman#2830: Modifies the same provider-config classifier to extend accepted rejection patterns (retrieved PR adds other remediation variants).
• tinyhumansai/openhuman#2464: Related changes to is_provider_config_rejection_message recognizing additional provider remediation shapes.
• tinyhumansai/openhuman#2309: Prior work adding new provider-config rejection substring arms and tests that overlap this classifier area.

Suggested labels

rust-core, sentry-traced-bug

Suggested reviewers

• graycyrus
• senamakel
• M3gA-Mind

"I nibble on strings of logs and tests,
sniff the phrases where a fallback rests,
a hop to route each noisy trace—
now watchers gather, tidy space,
the rabbit hums: the build’s at rest." 🐇

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The PR title accurately describes the main fix: demoting provider-config-rejection rollups in composio/triage error handling, which is the core objective of the changeset.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

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

---

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

[graycyrus]

LGTM — clean, surgical fix with good test coverage.

Nice work routing the composio triage error path through report_error_or_expected instead of the raw tracing::error!. The new phrase anchor in config_rejection.rs is well-scoped (sole producer at reliable.rs:332) and the cross-referencing comments will help catch drift.

Both changes follow the established pattern used across ~10 other call sites in the codebase. Tests cover the multi-line rollup and bare phrase cases. The {e:#} alternate Display for the error chain in the detail string is correct.

CI note: all failures are Docker login (GHCR auth) issues for the contributor's fork — unrelated to these changes. The Tauri build, E2E suites, and core image build all pass.

[coderabbitai]

Actionable comments posted: 0

[coderabbitai]

Actionable comments posted: 0

[M3gA-Mind]

PR #2689 — fix(composio/triage): demote provider-config-rejection rollups (Sentry TAURI-RUST-1V)

Walkthrough

Two surgical, additive changes that close a Sentry noise loop around [composio][triage] run_triage failed (10.7 k events / 14 d). The inner provider layer (reliable.rs + compatible.rs) already correctly classifies and demotes budget-exhausted / provider-config-rejection errors; the outer composio bus re-emitted the final aggregated error via a raw tracing::error! that bypassed the central classifier entirely. This PR routes that re-emit through report_error_or_expected and adds the canonical reliable-chain remediation phrase to the classifier so the full rollup message classifies on its way through. Both changes are consistent with the existing pattern for this class of Sentry noise (see #2481, #2612). No user-visible behaviour changes.

Changes

File	Summary
src/openhuman/inference/provider/config_rejection.rs	Adds "may not be available on your provider" to PHRASES; new test covers the full rollup body and the bare remediation phrase
src/openhuman/memory_sync/composio/bus.rs	Replaces raw tracing::error! at triage-failure site with report_error_or_expected; formats detail string to preserve label + anyhow error chain

Actionable comments (0)

None. The implementation is correct.

Verified / looks good

• Sole-producer claim confirmed. "may not be available on your provider" appears only in reliable.rs:324 (the producer) plus the classifier and its tests. The phrase is tight enough that no unrelated error would match it.
• report_error_or_expected call site is correct. Signature is (err: &E, domain, operation, extra: &[Tag<'_>]) where Tag<'a> = (&'a str, &'a str). The call passes detail.as_str() (full error chain via {e:#}), "composio", "trigger_triage", and &[("label", ...)]. The label field is preserved both inline in the detail string (visible in classified/demoted logs) and as a structured tag (used by Sentry for genuine errors). This mirrors how other bus handlers in the codebase call the classifier.
• apply_decision error path intentionally left as tracing::error!. That path fires when triage succeeds but the follow-on action fails — a genuine runtime error, not a provider-config rollup. Leaving it is correct.
• Tests. detects_reliable_chain_exhaustion_rollup covers the full multi-line rollup body (the TAURI-RUST-1V wire shape) and the bare single-line remediation phrase. Existing polarity tests (ignores_transient_and_server_and_unrelated, does_not_classify_unrelated_tools_phrases_as_config_rejection) continue to run unchanged.
• All 25 CI checks pass (Rust core tests, coverage ≥ 80%, E2E, fmt/clippy).
• No merge conflicts with current main.
• Project conventions followed: no new files at src/openhuman/ root, no dynamic imports, no secret logging, debug logging preserves the [composio][triage] grep-friendly prefix.