oxygen-agent.com

oxygen-diagnostics

Diagnose a running OXYGEN workspace: trace one cell or record value and its provenance, explain stuck or failed runs, sweep and retry table failures, inspect approvals and provider state, and explain or reduce credit spend.

First seen Jul 13, 2026

Installation

$ npx skills add https://oxygen-agent.com

Summary

  • Diagnose a running OXYGEN workspace: trace one cell or record value and its provenance, explain stuck or failed runs, sweep and retry table failures, inspect approvals and provider state, and explain or reduce credit spend.
  • Read-only investigation first; use the narrow playbook for the question.

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from oxygen-agent.com · top by installs.

npx skills add https://oxygen-agent.com

Browse all from oxygen-agent.com

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Skill metadata

Parsed from SKILL.md frontmatter.

Allowed toolsBash(oxygen *), Bash(oxygen-dev *), Bash(jq *)

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 1,744 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 36 installs

SKILL.md

OXYGEN Diagnostics

Choose one question; do not dump every log surface.

Question Read
Where did this value come from, or why is this cell stuck? [table-trace.md](table-trace.md)
Which rows/runs failed, and which are safe to retry? [error-sweep.md](error-sweep.md)
Why did credits increase, and how can spend be reduced? [optimize-credits.md](optimize-credits.md)

Use oxygen capabilities search "diagnose <symptom>" --json to identify Tables, Records, Workflows, Sequences, Agents, or Observability, then hydrate the exact read command/schema.

Evidence order

  1. Inspect the user-visible object and its deep-link.
  2. Read the owning run, item/cell state, provenance, provider, attempt, credits, and structured error.
  3. Correlate through Observability only as far as needed.
  4. Distinguish pending/leased/retryable/blocked/approval-required from a product bug.
  5. Preview a bounded retry or repair and explain whether it can spend credits or repeat an external write.
  6. Execute a retry only with the applicable approval; never retry an ambiguous external send without reconciliation.

If the evidence shows a general product defect, file it through oxygen-feedback with the relevant run/table/workflow IDs and transcript. Do not compensate with hidden local state or an alternate provider.