SuperPlane Skills

Agent skills for operating SuperPlane — the open source DevOps control plane for event-driven workflows.

Install

npx skills add superplanehq/skills

Or install a specific skill:

npx skills add superplanehq/skills --skill superplane-cli

Skills

Skill	Description
superplane-cli	Operate SuperPlane via CLI — auth, canvases, secrets, runs
superplane-canvas-builder	Design workflow canvases from requirements
superplane-monitor	Debug and inspect workflow executions

Running Evals

Regression tests for the skills. Each eval spawns a real Claude Code session with a skill loaded, gives it a task, and asserts the bash commands / canvas YAML / response Claude produces.

Run

export ANTHROPIC_API_KEY=sk-ant-...

make evals                              # all 15 cases
make evals CASES=push_to_slack          # one case
make evals SKILL=superplane-cli         # all cases for one skill
make evals.list                         # list case names without running

That's it. make evals builds the eval image, boots a fresh superplane-demo container alongside it on an internal Docker network, runs the cases, then tears the whole stack down.

Common knobs

EVAL_MODEL=claude-sonnet-4-5 make evals       # different model (default: claude-haiku-4-5)
EVAL_REPORT_FORMAT=markdown make evals        # markdown table only on stdout
EVAL_REPORT_FORMAT=text make evals            # superplane-style text report only
make evals.shell                              # bash inside the eval container for debugging
make evals.down                               # nuke leftover stack

Where to find results

After every run:

• evals/reports/<run_id>/report.md — per-case detail + summary table (open in any markdown viewer)
• evals/reports/<run_id>/summary.json — machine-readable totals + per-case stats
• evals/reports/<run_id>/<case_name>.json — full per-case detail (bash commands, files written, response text)
• tmp/evals/<run_id>-NN-<case_name>.log — timestamped event log per case