Applies named engineering principles and leaf skills to rigorous code work.
npx skills add https://github.com/Sma1lboy/rove --skill pstack-sma1lboy
Please help me install this Agent Skill. Skill: pstack Source: https://github.com/Sma1lboy/rove/tree/main/.agents/skills/pstack Command: npx skills add https://github.com/Sma1lboy/rove --skill pstack-sma1lboy
how (architecture explanation), why (design-rationale investigation across ADRs, git history, and MCP sources), interrogate (multi-model adversarial review), swarm (parallel worker fan-out), blast-radius (change impact analysis), and show-me-your-work (auditable TSV decision logs). - Use Case: Before shipping a contested refactor, run interrogate to have multiple LLM reviewers challenge the diff from independent angles, then synthesize a verdict categorized into act-on, consider, noted, and dismissed findings. ## Quick Start Ask the agent to go into pstack mode and apply rigorous principles to plan and verify your next nontrivial change.High-intent search queries and answers about installing and using this skill.
Invoke the interrogate skill, which packages the diff with a stated intent and rubric, spawns one reviewer per configured model in parallel, then synthesizes findings into act-on, consider, noted, and dismissed categories with an agreement map.
Use the why skill. It anchors the investigation in git blame and PR history, then fans out one investigator per evidence category (ADRs, design docs, issue trackers, chat, observability) and returns a cited, confidence-calibrated answer with gaps named.
The how skill explains what the code does: architecture, runtime flow, and file layout, with an optional multi-model critique mode. The why skill investigates motivation: the constraints, tickets, and discussions that forced the code's shape.
The 21 principle skills are pure prose and portable. Repo-specific parts, such as the local decision-record playbook and verification gates, reference Rove's ADRs, docs, and test commands, so other repos would need those references adapted.
Skip swarm for focused single-unit tasks that match a narrower playbook, and avoid over-fanning: parallelize only across genuine seams, give each worker its own worktree or branch, and declare race selection rules before spawning.
utsavanandGenerate unit tests, integration tests, or LLM eval artifacts for a feature.
kozyszooDetect, diagnose, and correct GitHub Actions CI failures on pull requests.
idchain-worldAutomate security-focused differential reviews of code changes with markdown reports.
Busboom-FooApply small code changes directly to master with a deterministic test workflow.