Argus

Forward-deployed expert · v4.2.1

When Argus surprises you, put an expert on it.

The argus-diagnostic skill is a senior Argus engineer you run inside your own agentic tool. Point it at a confusing run (an unexpected fix, a missing MR, a slow pipeline) and it reads the artifacts, tells you whether it’s working as designed or a genuine bug, and writes a report you can share with your team. It never leaves your environment.

Download the skillSee how it works

Ships with every release · versioned to the Argus binary you run.

Evidence, not guesses

Every conclusion points at a line in report.json, anchor.log, or your config. Never a hunch.

Knows this release cold

Deep knowledge of current behaviour: grand-slam-first qualification, the libraries and couplings policy, anchor and governor upgrades, and the telemetry triage flow, all versioned with the binary you run.

A report you can share

Summary → evidence → verdict → next steps, with secrets redacted, ready to share with your team.

The flow

Three steps

From “this looks wrong” to a filable report.

  • 01

    Prep

    Gather the artifacts. If you can re-run, do it with diagnostics turned up first. The anchor log (Argus’s record of every upgrade candidate it tried and why each was accepted or rejected) usually settles “why that version?”.

    report.json · config.yml · report-anchor.log · patches/ · bom.json · the full CI job log

  • 02

    Invoke

    Open your agentic tool in the folder with those files, name the skill, and describe what you saw versus what you expected. It reads the artifacts and asks any follow-ups.

    the dependency involved · the fix Argus chose · the fix you expected

  • 03

    Share

    It writes a self-contained report (summary, evidence, verdict, next steps) with secrets already redacted. Review it and share it with your team.

    working-as-designed vs candidate bug, with the exact artifact lines

terminal
# Re-run with full diagnostics (safe; --no-publish writes locally only)
ARGUS_PDF_REPORT=1 argus run \
  --pom pom.xml --config config.yml \
  --out patches --report report.json \
  --no-publish --log-level debug

Setup

Load it into your tool

The skill is a self-contained folder (SKILL.md plus references). Claude Code discovers it natively; for other tools you unpack it and point the agent at SKILL.md. The download is one argus-diagnostic-skill.zip file. Unzip it anywhere.

Claude Code
# Claude Code has native skills: just drop it in and it's auto-discovered.
unzip argus-diagnostic-skill.zip -d ~/.claude/skills/
# (or the project-local .claude/skills/ for a single repo)

# Then, in a folder holding your Argus artifacts, just ask. See "The prompt" below.

The ask

The prompt

Be concrete: name the coordinate, the strategy you saw, the strategy you expected, and which artifacts you’ve attached. The skill takes it from there.

Prompt
Use the argus-diagnostic skill. Argus opened an OVERRIDE for
tools.jackson.core:jackson-databind but I expected it to upgrade the json-utils
dependency instead. I've attached report.json, report-anchor.log, config.yml and
the CI job log. What happened, and is it a bug?

Act on it

Share the findings

The report is written to stand on its own. Review the redactions the skill applied (tokens, internal URLs, private coordinates), then share it with your team or security owner. A minimal reproducing pom.xml + config.yml alongside it turns a “maybe” into a fix.

Full reference

The skill runs entirely inside your agentic tool. Nothing is uploaded, and the report never leaves your environment unless you share it.