[Copilot]

Constraint Solving — Problem of the Day is pinned to claude-haiku-4.5, and threat detection inherits the pin. Switching to economy routing is estimated to cut usage from about 128 to 30–46 AIC/week and moves the workflow off the older model.

Workflow change (.github/workflows/constraint-solving-potd.md)

# removed: model: claude-haiku-4.5
engine:
  id: copilot
  bare: true
  model-routing:
    goal: cost
    mode: economy

Everything else is unchanged: prompt, triggers, cache-memory, evals, safe outputs and COPILOT_GITHUB_TOKEN authentication. There is no model allow/block list. Only this workflow was compiled; it is the only workflow diff against main.

Lock-file changes (constraint-solving-potd.lock.yml)

• Routing enabled: the agent's AWF config has experimental.modelRouting: true, the model config has routing: {enabled: true, goal: cost, mode: economy}, and the run step sets GH_AW_MODEL_ROUTING: 1.
• New step: "Prepare model-routing conversation".
• Router image: ghcr.io/github/gh-aw-firewall/router:0.29.6@sha256:fce2f3be… is pinned and downloaded.
• Pin removed: COPILOT_MODEL: claude-haiku-4.5 and the haiku cost table are gone from the agent job.
• Detection: now uses its default model (vars.GH_AW_MODEL_DETECTION_COPILOT || vars.GH_AW_DEFAULT_MODEL_COPILOT || 'detection', with an auto fallback).
• Downstream jobs: they read model, model_effort and model_routing_status from the agent job instead of the hard-coded model name.
• Artifacts: awf-routing-outcome.json and aw_info.json are added to the uploaded artifacts.
• Hashes: the frontmatter and admission hashes are regenerated.

Not in this PR

• The issue's "Check" step is a follow-up:
  • Audit the first routed runs with gh aw audit <run> against about 18 AIC/run.
  • Read the first week's discussions to confirm the models, constraints and pseudo-code are correct.
• To roll back, restore the model: claude-haiku-4.5 line.
• Most make agent-finish failures are on files this PR doesn't change: golint findings in pkg/cli, TestDownloadWorkflowLogsForTargets*, the JS tests send_otlp_span, awf_reflect_summary and copilot_sdk_mcp_config, and USE-001 in mcp_http_server_runner.cjs. validate-otel-contract, fix and security-govulncheck also fail and weren't investigated. None of these were re-run on main for comparison, and most are not on the known-failures list for main.

• Fixes Model routing: route Constraint Solving — Problem of the Day instead of pinning claude-haiku-4.5 #74588

[github-actions]

✅ PR Code Quality Reviewer completed the code quality review.

🔎 Code quality review by PR Code Quality Reviewer

[github-actions]

✅ Design Decision Gate 🏗️ completed the design decision gate check. See the comment below for the result and any generated ADR draft.

Design Decision Gate 🏗️

Outcome: reported_no_action

The agent completed and reported no action. A no-op output alone does not verify that all required work was completed or that publication was possible.

Agent messages are omitted from this comment. The category uses structured run status, not interpretation of agent prose.

Generated from Design Decision Gate 🏗️ · routed: gpt56 max · ~1 AIC · ⌖ ~2 AIC · ⊞ 7.7K · ◷

🏗️ ADR gate enforced by Design Decision Gate 🏗️

[github-actions]

✅ Ponytail Reviewer completed successfully!

Ponytail Reviewer

Outcome: reported_no_action

The agent completed and reported no action. A no-op output alone does not verify that all required work was completed or that publication was possible.

Agent messages are omitted from this comment. The category uses structured run status, not interpretation of agent prose.

Generated from Ponytail Reviewer · routed: gpt56 max · ~1 AIC · ⌖ ~1 AIC · ⊞ 118 · ◷

Generated by Ponytail Reviewer for #74595

[github-actions]

✅ Test Quality Sentinel completed test quality analysis.

Test Quality Sentinel

Outcome: reported_no_action

The agent completed and reported no action. A no-op output alone does not verify that all required work was completed or that publication was possible.

Agent messages are omitted from this comment. The category uses structured run status, not interpretation of agent prose.

Generated from Test Quality Sentinel · sonnet50 · ~16 AIC · ⌖ ~21 AIC · ⊞ 12.9K · ◷

🧪 Test quality analysis by Test Quality Sentinel

[github-actions]

🧠 Matt Pocock Skills Reviewer has completed the skills-based review. ✅

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer

[Copilot]

🟢 Approval recommended

No active review issues are reported for the switch to cost-focused economy routing.

No open issues 🎉

What changed in this PR

Updates the daily constraint-solving workflow to use cost-focused economy model routing instead of a fixed claude-haiku-4.5 model, implementing the configuration change requested in #74588.

Changes:

• Enables economy routing while preserving Copilot CLI, bare: true, authentication, prompts, and workflow controls.
• Regenerates the lock file with router setup, routed-model reporting, and routing artifacts; threat detection returns to default model selection.
• Leaves cost auditing and output-quality checks for follow-up; projected savings remain unverified by routed runs.

File	Description
.github/​workflows/​constraint-solving-potd.md	Replaces the fixed model with cost-focused economy routing.
.github/​workflows/​constraint-solving-potd.lock.yml	Adds compiled routing support, updates model reporting, and removes the inherited detection pin.

_{Review effort: Balanced (auto)}

[github-actions]

Reviewed the changed workflow configuration and generated-lock implications. I found no actionable correctness, security, or maintainability issues on the changed lines. The economy routing configuration matches existing repository patterns; generated lock-file changes are represented in PR metadata though excluded from the capped review patch.

🔎 Code quality review by PR Code Quality Reviewer · copilot · gpt50 · ~1 AIC · ⌖ ~1 AIC · ⊞ 1.6K
_{Comment /review to run again}

[github-actions]

Skills-Based Review 🧠

Applied /grill-with-docs to the workflow change. The routing config is valid and the committed lock matches the description. One behaviour change needs confirming before merge.

📋 Key Themes & Highlights

Key Themes

• Threat detection is unpinned: removing model: claude-haiku-4.5 also removes the model that detection inherited. Detection now falls back to the detection alias (small → mini), which can resolve to a different model. Detection is a security gate, so this should be pinned or called out explicitly.
• PR description mismatch: it says detection falls back to auto, but the committed lock uses 'detection'.

Positive Highlights

• ✅ The model-routing block matches the schema (goal: cost, mode: economy) and other routed workflows in the repo.
• ✅ The lock adds the routing step, AWF modelRouting config, and the pinned router image, consistent with the source.

Scope

Reviewed committed head 65365ec. The lock file is generated, so I checked only its routing-related lines.

@copilot please address the inline comment on threat-detection model handling.

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer · copilot · routed: haiku55 xhigh · ~6 AIC · ⌖ ~1 AIC · ⊞ 5.5K
_{Comment /matt to run again}