powernode-platform: Skill for Claude Code

/campaign — Autonomous Improvement Campaigns

A campaign is a durable, named, repeatable improvement run. It replaces the hand-authored standing prompt + ~/.claude plan files: a run becomes start(config). The campaign owns its scope/posture/decision-authority/stop-conditions, the Ralph loop(s) it drives, a decision log, an async parked-questions queue, and a progress ledger. /campaign run is the Ralph-pattern loop body — drive it repeatedly with /loop /campaign run <id>.

Dual surface — platform agents call the same actions through the MCP tools platform.campaign_start, platform.campaign_status, platform.campaign_answer_question, platform.campaign_stop and platform.campaign_resume.

Usage

/campaign start <name>     # create a campaign + its dedicated loop (interactive config)
/campaign run [<id>]       # drive ONE improvement iteration for the campaign (loop body)
/campaign status [<id>]    # progress + open questions + recent decisions (all, or one)
/campaign answer <id>      # answer a parked question to unblock the campaign
/campaign stop <id>        # stop the campaign + pause its loops + record a summary
/campaign resume <id>      # request a resume (human-confirmed; parks until a person approves)

start

Surface the config, then create. Confirm these with the operator if not given:

• scope — which tree/area (e.g. a private extension, server/, worker/); stays in configuration.
• posture/ordering — bugs-first, dead-code, test-gaps, quality; free text in configuration.
• decision_authority — how much the loop decides vs parks (see table below). Default trusted.
• stop_conditions — e.g. { "max_failed": 3, "completion_pct": 100 }. Optional.

Call platform.campaign_start(name:, description:, decision_authority:, configuration:, stop_conditions:). It creates the Ai::Campaign and a dedicated campaign-scoped Ralph loop (branch campaign/<id>), starts the campaign, and snapshots initial progress. Report the campaign id + loop branch.