Skip to content

[agent-efficiency] Recurring agent workflow failures: checkout ref errors and missing agent-output artifact #419

Description

@github-actions

Scope

Lookback window: 2026-02-19T17:11:13Z to 2026-02-26T17:11:13Z (last 7 days).

Data analyzed:

  • /tmp/gh-aw/failed_runs.json
  • /tmp/gh-aw/errors.json
  • Agent/copilot step logs under /tmp/gh-aw/logs/(run_id)/
  • Workflow run metadata from GitHub Actions run listings (filtered to paths matching trigger-|gh-aw-)

Per-repository run summary (last 7 days)

Repository Runs (trigger/gh-aw paths) Completed Success Failure Pass rate Fail rate Cancelled Skipped Action required
elastic/ai-github-actions 42 42 8 3 19.0% 7.1% 4 20 2
elastic/ai-github-actions-playground 0 0 0 0 0.0% 0.0% 0 0 0
elastic/beats 39 39 0 0 0.0% 0.0% 0 39 0
elastic/integrations 0 0 0 0 0.0% 0.0% 0 0 0

Behavioral patterns

1) Checkout strategy repeatedly fails on missing remote PR head refs

What happened
Runs failed during agent checkout with git fetch origin (pr-head-branch) followed by missing remote ref.

How many runs / workflows

  • 11 failed runs in this window
  • Workflows: .github/workflows/trigger-update-pr-body.yml (6), .github/workflows/trigger-pr-review.yml (5)

Exact evidence

Root cause (observed fragment)
The failing path is the pull_request checkout strategy in agent execution (Strategy: git fetch + checkout in the same log block at lines 985-989), which depends on PR head refs being fetchable by name at run time.

2) Conclusion/safe-output flow repeatedly attempts to download missing agent-output artifact

What happened
After upstream failures, follow-up steps repeatedly emit Artifact not found for name: agent-output.

How many runs / workflows

  • 12 failed runs in this window
  • Workflows: .github/workflows/trigger-update-pr-body.yml (7), .github/workflows/trigger-pr-review.yml (5)

Exact evidence

Root cause (observed file fragment)
These workflows include a download step for agent-output:

  • .github/workflows/gh-aw-update-pr-body.lock.yml:1296-1301
  • .github/workflows/gh-aw-pr-review.lock.yml:1469-1471

3) Model/endpoint incompatibility observed in a failed run

What happened
Agent execution failed after retries with a model endpoint compatibility error.

How many runs / workflows

  • 1 failed run in this window
  • Workflow: .github/workflows/trigger-update-pr-body.yml

Exact evidence

Root cause (observed file fragment)
The workflow source sets the default model to gpt-5.3-codex:

  • .github/workflows/gh-aw-update-pr-body.md:21-25

Notes on requested behavior analysis dimensions

  • No concrete evidence of excessive tool-call loops was found in the available failed-run agent logs for this window.
  • No concrete evidence in these logs of MCP bulk-query misuse by the agent within run prompts.

Existing related issue found during duplicate check: #389 (open). This report covers the current 7-day window and includes updated counts plus per-repository run metadata.

Generated by Internal: Agent Efficiency

  • expires on Mar 5, 2026, 5:15 PM UTC

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    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