[agent-deep-dive] Agent deep dive: gh-aw-branch-actions-detective.lock.yml - 2026-11-18

Aperta
#2,530 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
3/5
Tempo stimato
1-2 giorni
Idoneità per principianti
48/100
Tipo di issue
Bug
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
github-actions
Ambito
devops

Direzione di ricerca

Start by reading the agent-deep-dive workflow prompt and the workflow or preprocessing steps that fetch job metadata and run logs; the issue does not name their files. Trace what happens when the jobs list is empty or log download returns 404, then compare the handling with the suggested missing-data paths. Done when unavailable logs produce a clear, non-failing outcome with useful run metadata.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

agentic-workflows

Agent Deep Dive: gh-aw-branch-actions-detective.lock.yml

Date: 2026-11-18
Runs analyzed: 6 total (0 success, 6 failure, 0 other)

Run Outcomes

Run Date Conclusion
29840729033 2026-07-21 failure
29836290985 2026-07-21 failure
29836119982 2026-07-21 failure
23928110386 2026-04-03 failure
23888698120 2026-04-02 failure
23862510988 2026-04-01 failure

Failure Analysis

For every failed run:

  • Step name: unavailable (job list API returns no jobs)
  • Error excerpt: failed to get workflow run logs: unexpected status code: 404 Not Found
  • Additional evidence: preloaded /tmp/gh-aw/deep-dive/logs/<run_id>/logs.zip files contain JSON {"message":"Not Found","status":"404"} instead of zip archives
  • Likely cause: the workflow (or preprocessing pipeline) attempts to analyze runs whose job/log artifacts are no longer retrievable (retention expiry or inaccessible run-log endpoint), and does not gracefully degrade when logs are missing.

Tool Call Patterns

  • Average turns per run: not measurable from available artifacts (agent step logs unavailable)

  • Most-called tools: not measurable from available artifacts

  • **Anomalies observed in available (redacted)

    • 6/6 runs have conclusion: failure
    • 6/6 runs return total_count: 0 from workflow jobs listing
    • 6/6 runs return 404 for workflow log download endpoint

Behavioral Observations

  • Intended behavior from workflow prompt: fetch failed jobs and logs, analyze root cause, emit create_issue or noop.
  • Observed behavior from run artifacts: no run had retrievable jobs/logs, so required analysis inputs were missing in every case.
  • Because success examples are absent, output-quality evaluation for successful runs cannot be performed.

Recurring Patterns

Patterns seen in 2+ runs:

  1. Every run is a failure (6/6).
  2. Every run has missing job metadata (jobs: [], total_count: 0).
  3. Every run has missing logs (404 from log endpoint).
  4. The failure pattern is stable across two separate time windows (April and July), indicating a systemic reliability gap rather than a one-off outage.

Prompt Improvement Suggestions

  1. Add explicit missing-log handling path in prompt.

    • Observed problem: log retrieval returns 404 for all analyzed runs.
    • Prompt change: instruct the agent to treat 404/410 log responses as a terminal non-actionable condition and emit noop with run ID + reason, rather than continuing to failing paths.
    • Expected improvement: fewer hard-failure runs; clearer operator signal when data retention prevents analysis.
  2. Require pre-check of job availability before log download.

    • Observed problem: all runs have jobs: [], yet logs are still attempted.
    • Prompt change: if jobs list is empty, short-circuit with a dedicated message (No jobs available for run (likely expired)), skip deeper analysis.
    • Expected improvement: fewer unnecessary calls; faster completion; reduced error noise.
  3. Constrain analysis window to retrievable runs.

    • Observed problem: analyzed runs are historical and all missing logs/jobs.
    • Prompt change: prioritize recent runs within retention window and/or verify log URL accessibility before selecting runs to analyze.
    • Expected improvement: higher chance of actionable root-cause extraction; better issue quality.
  4. Add fallback evidence mode.

    • Observed problem: when logs are unavailable, current workflow yields little diagnostic value.
    • Prompt change: direct agent to capture available metadata only (run/event/branch/head SHA/failure recurrence) and produce a concise infrastructure/data-availability issue.
    • Expected improvement: consistent, useful output even when deep logs are unavailable.

Generated by Internal: Agent Deep Dive · copilot · gpt53codex · 39.4 AIC · ⌖ 5.66 AIC · ⊞ 25.1K · ◷

  • expires on Dec 2, 2026, 3:01 PM UTC
Lingua principale
Python
Stelle
13
Fork
16
Merge medio
45m
PR unite (30g)
1

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di elastic/ai-github-actions

Tutte le issue di elastic/ai-github-actions

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.