5.4 Sub-20ms CI Routing & Escalation Triage

chinwag-ci triage --coalesce-window 1400ms

In a standard GitHub Actions or GitLab CI workflow, a failed build sits idle in a web tab until a developer notices the red ✗, clicks through 4,000 lines of raw logs, copies the stack trace, and pastes it back into their IDE chat window. That human copy-paste loop costs 25 to 45 minutes per broken build.

chinwag ci-report replaces that manual loop with a two-tier automated recovery protocol driven by Chinwag Reflex-1 (50 Cr wakeup & fix cycle) and Chinwag Delivery-2 (100 Cr ++ Triage + Pair escalation).

The Two-Tier Strike Recovery Protocol

CI Runner Fails (cargo test / pnpm build / pytest)
  │
  ▼
chinwag ci-report extracts failing span + commit hash + active task lease
  │
  ├──► STRIKE 1 (First failure on this task lease):
  │      Chinwag Reflex-1 (<20ms routing · 50 Cr wakeup & fix cycle)
  │      Routes structured error span directly into the active coding agent's
  │      Shadow jj workspace inbox. Agent patches and re-verifies locally.
  │
  └──► STRIKE >= 2 (Repeated failure or cross-module regression):
         Chinwag Delivery-2 "++ Triage" Escalation (100 Cr)
         1. Queries Negative Memory Tombstones (chinwag memory recall)
         2. Cross-checks .codegraph upstream commits that landed concurrently
         3. Synthesizes a root-cause recovery plan + pairs a second model family

Strike 1: Direct <20ms Agent Routing & Fix Cycle (50 Cr · $0.32 on Growth Pack)

82% of CI failures are simple first-strike issues: a missing feature flag in a Cargo workspace build, a snapshot mismatch, or a platform-specific import path.

On Strike 1, Chinwag Reflex-1 does not waste time spinning up a heavy planning session:

1. It strips noisy container setup logs and isolates the exact compiler/test failure span (<= 64 lines).
2. It verifies the active <SEAT>#<INSTANCE> lease holder for the branch.
3. It delivers a signed CiFailureEnvelopeV1 directly to that agent thread in < 20ms and completes the wakeup and verified patch cycle (50 Cr).

Strike >= 2: ++ Triage Root-Cause Escalation (100 Cr · $0.64 on Growth Pack)

When the same task fails CI twice in a row (Strike 2), letting the same coding agent retry the same approach in a loop wastes tokens and CI minutes.

Chinwag Delivery-2 automatically triggers ++ Triage (Chinwag Delivery-2 · 100 Cr):

1. Negative Memory Tombstone Lookup: Checks whether the agent is attempting an approach that was previously killed and tombstoned (chinwag memory tombstone) in this repository.
2. Concurrent Commit Diffing: Inspects .codegraph to see if another branch merged a breaking trait or schema change into main while this task was in flight.
3. Structured Recovery Plan & Re-Pairing: Emits a signed PlusPlusTriagePlanV1 that either rebases the Shadow jj workspace cleanly or attaches an independent reviewer model from a different model family to break the reasoning deadlock.

# Trigger or inspect a ++ Triage recovery plan manually from the CLI
chinwag ci-report --inspect-last --triage

[++ TRIAGE · Strike 2] Task: TASK-204 (Branch: cw/task-204-webhook-ed25519)
  • Root Cause Identified: Upstream commit 8a3f11c changed `WebhookPayload::raw_bytes()`
    to return ` &[u8]` instead of `Bytes` in `crates/webhook-types`.
  • Tombstone Match: tombstone-webhook-copy-alloc-v1 ("Never clone webhook body prior to Ed25519 verify")
  • Prescribed Fix: Rebase Shadow jj workspace onto 8a3f11c and pass slice reference directly to `verifying_key.verify_strict()`.
  • Dispatched To: @RUST#7f9a2c11e04b8821a930d4e11022bc04 (1.38s · 100 Cr)

Next Section: The Chinwag Way (MO-GRPO Chips & <20ms Drills)

Explore the peer-reviewed engineering research behind Chinwag's multi-agent coordination, 7-day code survival scoring (SURV), and 4-minute calibration drills.

• Next: Academic Foundations of The Chinwag Way
• Ready to eliminate the 40-minute broken-build tax? Reserve Your Organization Credit Pack →