Skip to content

[agent-efficiency] Recent agent runs show AI-credit and uv setup failure clusters #1454

Description

@github-actions

Run summary

Lookback window: last 7 days ending 2026-06-29T17:52:20Z. The setup provided /tmp/gh-aw/failed_runs.json with 20 failed runs and downloaded logs for those runs. A GitHub MCP run-list page covering the most recent 30 workflow runs back to 2026-06-28T10:41:50Z showed 20 matching trigger-/gh-aw- runs; full 7-day pass/fail pagination was not available in this sandbox. Downstream discovery with GitHub code search failed with a 429, so no downstream repository rows were discovered in this run.

Repository Metadata source Total observed matching runs Success Failure Other Pass rate Fail rate
elastic/ai-github-actions GitHub MCP page 1, recent runs through 2026-06-28T10:41:50Z 20 11 8 1 queued 57.9% 42.1%
elastic/ai-github-actions Predownloaded failed-run corpus for the lookback window 20 failed runs n/a 20 n/a n/a n/a

Open issue duplicate check found individual per-run failure issues such as #1447, #1451, #1445, #1448, #1446, and #1444, plus prior aggregate reports #1433 and #1407. This report groups the fresh pattern observed in the currently downloaded run corpus.

Findings

1. Seven downloaded runs hit the AI credits guardrail

What happened: 7 of the 20 downloaded failed runs ended with AI credits rate-limit error: true in conclusion logs.

Run Workflow Evidence
https://github.com/elastic/ai-github-actions/actions/runs/28241973065 Text Auditor /tmp/gh-aw/logs/28241973065/1_run _ conclusion.txt: GH_AW_AI_CREDITS_RATE_LIMIT_ERROR: true; [ai-credits] rateLimitSignal source=env(GH_AW_AI_CREDITS_RATE_LIMIT_ERROR); AI credits rate-limit error: true
https://github.com/elastic/ai-github-actions/actions/runs/28242001792 Framework Best Practices /tmp/gh-aw/logs/28242001792/1_run _ conclusion.txt: GH_AW_AI_CREDITS_RATE_LIMIT_ERROR: true; [ai-credits] rateLimitSignal source=env(GH_AW_AI_CREDITS_RATE_LIMIT_ERROR); AI credits rate-limit error: true
https://github.com/elastic/ai-github-actions/actions/runs/28252173493 Autonomy Atomicity Analyzer /tmp/gh-aw/logs/28252173493/0_run _ conclusion.txt: GH_AW_AI_CREDITS_RATE_LIMIT_ERROR: true; [ai-credits] rateLimitSignal source=env(GH_AW_AI_CREDITS_RATE_LIMIT_ERROR); AI credits rate-limit error: true; Unknown model AI credits error: true
https://github.com/elastic/ai-github-actions/actions/runs/28376490735 Agent Suggestions /tmp/gh-aw/logs/28376490735/0_run _ conclusion.txt: [ai-credits] rateLimitSignal source=env(GH_AW_AI_CREDITS_RATE_LIMIT_ERROR); AI credits rate-limit error: true; Unknown model AI credits error: true
https://github.com/elastic/ai-github-actions/actions/runs/28379246071 Text Auditor /tmp/gh-aw/logs/28379246071/1_run _ conclusion.txt: GH_AW_AI_CREDITS_RATE_LIMIT_ERROR: true; [ai-credits] rateLimitSignal source=env(GH_AW_AI_CREDITS_RATE_LIMIT_ERROR); AI credits rate-limit error: true
https://github.com/elastic/ai-github-actions/actions/runs/28379277454 Framework Best Practices /tmp/gh-aw/logs/28379277454/1_run _ conclusion.txt: GH_AW_AI_CREDITS_RATE_LIMIT_ERROR: true; [ai-credits] rateLimitSignal source=env(GH_AW_AI_CREDITS_RATE_LIMIT_ERROR); AI credits rate-limit error: true
https://github.com/elastic/ai-github-actions/actions/runs/28382426387 Refactor Opportunist /tmp/gh-aw/logs/28382426387/0_run _ conclusion.txt: [ai-credits] rateLimitSignal source=env(GH_AW_AI_CREDITS_RATE_LIMIT_ERROR); AI credits rate-limit error: true; Unknown model AI credits error: true

Affected workflows: 5 workflows: gh-aw-text-auditor.md (2 runs), gh-aw-framework-best-practices.md (2), gh-aw-autonomy-atomicity-analyzer.md (1), gh-aw-agent-suggestions.md (1), and gh-aw-refactor-opportunist.md (1).

Root cause evidence: These workflows all default to gpt-5.3-codex and allow 90-minute runs. Four also explicitly require Pick Three fanout with three general-purpose sub-agents. Relevant prompt/config locations:

Workflow Config / prompt evidence
.github/workflows/gh-aw-text-auditor.md lines 24-28 default gpt-5.3-codex; line 110 timeout-minutes: 90; lines 121-123 broad text-audit assignment
.github/workflows/gh-aw-framework-best-practices.md lines 25-29 default gpt-5.3-codex; line 89 timeout-minutes: 90; lines 121-124 require three general-purpose sub-agents
.github/workflows/gh-aw-agent-suggestions.md lines 24-28 default gpt-5.3-codex; line 85 timeout-minutes: 90; lines 111-120 require three general-purpose sub-agents plus downstream review
.github/workflows/gh-aw-refactor-opportunist.md lines 25-29 default gpt-5.3-codex; line 85 timeout-minutes: 90; lines 108-113 require three general-purpose sub-agents plus churn/issues review
.github/workflows/gh-aw-autonomy-atomicity-analyzer.md lines 24-28 default gpt-5.3-codex; line 82 timeout-minutes: 90; lines 107-114 require broad scans and three general-purpose sub-agents

2. Four code-quality detector runs failed before the agent could start because UV_PATH was empty

What happened: 4 of the 20 downloaded failed runs failed in setup while trying to expose uv. The agent logs show UV_PATH: was empty, then ln attempted to create the symlink target from an empty source and exited 1.

Run Workflow Evidence
https://github.com/elastic/ai-github-actions/actions/runs/28379087056 Code Complexity Detector /tmp/gh-aw/logs/28379087056/4_run _ agent.txt: UV_PATH: followed by ln: failed to create symbolic link '/opt/hostedtoolcache/gh-aw-tools/current/x64/bin/uv' -> '': No such file or directory and Process completed with exit code 1.
https://github.com/elastic/ai-github-actions/actions/runs/28375501681 Code Duplication Detector /tmp/gh-aw/logs/28375501681/4_run _ agent.txt: same empty UV_PATH, failed symlink, and exit code 1
https://github.com/elastic/ai-github-actions/actions/runs/28241805123 Code Complexity Detector /tmp/gh-aw/logs/28241805123/4_run _ agent.txt: same empty UV_PATH, failed symlink, and exit code 1
https://github.com/elastic/ai-github-actions/actions/runs/28238693205 Code Duplication Detector /tmp/gh-aw/logs/28238693205/4_run _ agent.txt: same empty UV_PATH, failed symlink, and exit code 1

Affected workflows: 2 workflows: gh-aw-code-complexity-detector.md (2 runs) and gh-aw-code-duplication-detector.md (2 runs).

Root cause evidence: Both detector workflows define a Serena MCP server using uvx; comments state setup-uv is emitted by the compiler (.github/workflows/gh-aw-code-complexity-detector.md lines 81-95 and .github/workflows/gh-aw-code-duplication-detector.md lines 81-95). The shared hand-authored setup fragment then uses the setup-uv output directly: .github/workflows/gh-aw-fragments/runtime-setup.md lines 35-50 runs astral-sh/setup-uv@v5, sets UV_PATH: ${{ steps.setup-uv.outputs.uv-path }}, and executes sudo ln -sf "$UV_PATH" "$toolcache_bin/uv". In the failed logs, that output was empty.

Other observed failed runs

The remaining downloaded failed runs did not show the same two patterns in the inspected logs. Two downloaded entries only contained invalid logs.zip placeholders (28224970412, 28224970824), and the rest either completed agent execution or were individual non-pattern failures already reported by the workflow failure reporter.

Generated by Internal: Agent Efficiency · 643.8 AIC · ⌖ 22 AIC · ⊞ 24.9K ·

  • expires on Jul 6, 2026, 5:58 PM UTC

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions