QA + Agent Triage Pipeline#

This showcase targets maintainers of the host library running QA and AI-assisted triage.

Problem#

You need fast, reproducible diagnosis across many failures while keeping human review as the final decision point.

Pattern#

Run strict diagnostics in CI, export structured bundles, and triage by stable keys (code, signal, trace_hash).

import smonitor

smonitor.configure(
    profile="qa",
    strict_signals=True,
    strict_schema=True,
    buffer_events=True,
)

pytest
smonitor export --out smonitor_bundle --max-events 500

Automation pipeline:

  1. CI stores the bundle as an artifact.

  2. Agent parses event stream in agent profile.

  3. Triage groups incidents by code + trace_hash.

  4. Agent proposes diagnosis and patch draft.

  5. Maintainer reviews and merges only after tests pass.

Deterministic triage key#

For repeated incidents across runs, derive a stable key from:

  • code

  • source

  • selected context fields (for example library, version)

import hashlib
import json


def triage_key(event):
    context = event.get("context") or {}
    signature = {
        "code": event.get("code"),
        "source": event.get("source"),
        "library": context.get("library"),
        "version": context.get("version"),
    }
    digest = hashlib.sha256(
        json.dumps(signature, sort_keys=True, separators=(",", ":")).encode("utf-8")
    ).hexdigest()[:16]
    return f"{signature['code']}:{digest}"

This avoids brittle grouping by free text while keeping incidents comparable.

Why this works#

  • failures become comparable across runs and environments;

  • triage focuses on stable contracts, not brittle free text;

  • human ownership is preserved.

Where to apply#

  • nightly test pipelines;

  • release-candidate validation;

  • large issue backlogs with repeated incident classes.