Skip to content

Route Documentation Unbloat instead of pinning claude-sonnet-5.5 - #73388

Merged
SivaKesava1 merged 2 commits into
mainfrom
copilot/model-routing-route-documentation-unbloat
Oct 31, 2026
Merged

SivaKesava1 merged 2 commits into
mainfrom
copilot/model-routing-route-documentation-unbloat

Conversation

Copilot AI commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

Documentation Unbloat pins copilot/claude-sonnet-5.5, so threat detection and evals also run on Sonnet. Those two jobs account for about 71% of its roughly 28 AIC per run. This PR replaces the pin with model routing (goal: cost, mode: economy), following the earlier routed workflows.

Workflow frontmatter (unbloat-docs.md)

engine:
  id: pi
  model-provider: github
  model-routing:
    goal: cost
    mode: economy
  • Removes model: copilot/claude-sonnet-5.5.
  • max-turns, the prompt, triggers, tools, sub-agents and safe outputs are unchanged.
  • No allowed-models or blocked models.

Compiled lock (unbloat-docs.lock.yml)

  • Pins the AWF router image ghcr.io/github/gh-aw-firewall/router:0.29.0@sha256:….
  • Adds a "Prepare model-routing conversation" step.
  • Enables model routing in the agent's AWF config, with no allowed, blocked or candidate models.
  • Removes GH_AW_PI_MODEL and GH_AW_PI_NATIVE_PROVIDER from the agent step.
  • Passes the routed model, effort and routing status (model, model_effort, model_routing_status) to the downstream jobs.
  • Threat detection and evals use their detection and evals default models instead of the pin.

make recompile changed no other workflows.

Validation

make agent-finish stopped at test-integration, so its later steps didn't run. The only failure is TestNoRawOctalPermissionLiteralsInOSCalls, which flags pkg/cli/logs_harbor_trial.go. This PR doesn't touch that file, so it should also fail on main; that hasn't been checked. The CodeQL scan timed out and didn't complete.

Follow-up (not in this PR)

The issue's post-merge checks:

  • Audit the first routed runs. The issue expects mostly gpt-5.6-luna at max effort.
  • Compare AIC per run across the agent, detection and evals jobs with about 28 today.
  • Compare the PR merge rate with 83%. If it drops, switch to mode: balanced.
…ude-sonnet-5.5

Co-authored-by: SivaKesava1 <11771739+SivaKesava1@users.noreply.github.com>
Copilot AI changed the title [WIP] Update routing for Documentation Unbloat to use model routing Oct 31, 2026
Copilot AI requested a review from SivaKesava1 October 31, 2026 02:27
@SivaKesava1
SivaKesava1 marked this pull request as ready for review October 31, 2026 02:28
Copilot AI auto review requested due to automatic review settings October 31, 2026 02:28
@github-actions

github-actions Bot commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

✅ 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 · gpt56 · ~7 AIC · ⌖ ~6 AIC · ⊞ 15.2K · ◷

Generated by Ponytail Reviewer for #73388

@github-actions

github-actions Bot commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

✅ 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 · haiku45 · ~6 AIC · ⌖ ~7 AIC · ◇ ~3 AIC · ⊞ 9.3K · ◷

🧪 Test quality analysis by Test Quality Sentinel

@github-actions

github-actions Bot commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

✅ 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 🏗️ · opus55 · ~11 AIC · ⌖ ~33 AIC · ◇ ~1 AIC · ⊞ 2.0K · ◷

🏗️ ADR gate enforced by Design Decision Gate 🏗️

@github-actions

github-actions Bot commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

✅ PR Code Quality Reviewer completed the code quality review.

🔎 Code quality review by PR Code Quality Reviewer

@github-actions

github-actions Bot commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

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

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

@github-actions

Copy link
Copy Markdown
Contributor

Comment Memory

Peek at saved memory (pr-code-quality-reviewer)
reviewed_at: 2026-10-31T02:31:18Z
review_event: COMMENT
top_themes:
  - unconstrained cost/economy model routing may make scheduled PR quality nondeterministic
files_reviewed:
  - .github/workflows/unbloat-docs.md
  - .github/workflows/unbloat-docs.lock.yml (runtime details noted but generated patch was excluded)
comment_count: 1

Note

This comment is managed by comment memory.

Expand the saved memory block to view or edit the persistent context for this thread.
Edit only the text inside the backtick fences; workflow metadata and the footer are regenerated automatically.

Learn more about comment memory

🔎 Code quality review by PR Code Quality Reviewer · copilot · gpt50 · ~2 AIC · ⌖ ~1 AIC · ⊞ 1.7K · ◷
Comment /review to run again

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review: comment

One medium-risk concern was identified: cost/economy routing is unconstrained, so the scheduled PR-producing workflow can change behavior as the provider model pool changes. See the inline comment.

💡 Review scope

Reviewed the changed workflow frontmatter and the supplied PR metadata. The generated lock file was not included in the pre-fetched patch, so its runtime details were not independently assessed. No additional blocking correctness issue was established from the available changed lines.

🔎 Code quality review by PR Code Quality Reviewer · copilot · gpt50 · ~2 AIC · ⌖ ~1 AIC · ⊞ 1.7K
Comment /review to run again

Comment thread .github/workflows/unbloat-docs.md Resolved

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟢 Approval recommended

There are no active review issues for the cost-focused model-routing change.

No open issues 🎉

What changed in this PR

Moves Documentation Unbloat to gh-aw’s cost-focused model routing, following the rollout in #73381. The change aims to reduce run costs without changing the documentation task or its safeguards.

Changes:

  • Replaces the Sonnet pin with goal: cost and mode: economy, retaining Pi and the 90-turn limit.
  • Adds router setup and passes the selected model, effort, and routing status downstream.
  • Switches threat detection and evals to their default model aliases.
File Description
.github/​workflows/​unbloat-docs.md Enables cost/economy routing; leaves the prompt, tools, and safe outputs unchanged.
.github/​workflows/​unbloat-docs.lock.yml Compiles routing setup, pinned router image, runtime metadata, and downstream model defaults.

Review effort: Balanced (auto)

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Skills-Based Review 🧠

Applied /codebase-design (pr-triage recommended this + /improve-codebase-architecture). The change mechanically follows the established model-routing pattern used across the repo (copilot-session-insights.md, safe-output-health.md, claude-code-user-docs-review.md all use goal: cost / mode: economy) — one worthwhile mode question flagged inline.

📋 Key Themes & Highlights

Key Theme

  • Mode choice vs. workflow risk profile: unbloat-docs.md creates PRs with auto-merge: true — the project's own model-routing.md docs recommend mode: robust for exactly this class of workflow ("tasks where a failed run is costly, such as workflows that open pull requests"), while economy is documented for "triage, labeling, and summaries." Sibling routed workflows that auto-merge or act with less oversight (avenger.md) use mode: balanced, not economy. Worth a deliberate justification in the PR description if economy is intentional, given the auto-merge safety net is thinner than human-reviewed PR flows.

Positive Highlights

  • ✅ Change is minimal and consistent with the repo's existing model-routing migration pattern (follows the same engine.model-routing shape used in ~10 other workflows).
  • ✅ PR description is unusually thorough: cites the AIC math driving the change, lists exactly what gh aw compile altered in the lock file, and calls out validation gaps (CodeQL timeout, pre-existing TestNoRawOctalPermissionLiteralsInOSCalls failure) honestly instead of hiding them.
  • ✅ Correctly scoped — max-turns, prompt, triggers, tools, and safe-outputs are untouched; only the model selection strategy changes.
  • ✅ Lock file regeneration via gh aw compile was confirmed to not affect other workflows.

Note on investigation scope: This review relied on the pre-fetched 18-line diff (small, not truncated) plus targeted doc/sibling-workflow reads; no cross-cutting architectural exploration was needed given the PR's narrow footprint.

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer · copilot · sonnet50 · ~63 AIC · ⌖ ~16 AIC · ⊞ 10.8K
Comment /matt to run again

Comment thread .github/workflows/unbloat-docs.md Resolved
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

3 participants