Scores implementation plans against a rubric and iterates them before coding.
npx skills add https://github.com/sicambria/talkteach-asr --skill planmax-sicambria
Please help me install this Agent Skill. Skill: planmax Source: https://github.com/sicambria/talkteach-asr/tree/main/.claude/skills/planmax Command: npx skills add https://github.com/sicambria/talkteach-asr --skill planmax-sicambria
.harness/config.json for plan directories, dimensions, and verification commands, degrading gracefully to generic templates when absent. - Use Case: Before implementing a non-trivial feature, invoke the Skill to draft a plan citing concrete path:line evidence, iterate it past a 90-point threshold with advisor approval, then implement following the repo's operating contract. ## Quick Start Ask the assistant to run planmax on your task, for example by saying "plan it out and score it" for the feature you want to implement.High-intent search queries and answers about installing and using this skill.
Invoke planmax with your task and it drafts or reuses a plan, then scores it 0-100 across six weighted axes including evidence grounding, concreteness, and risk. It revises the lowest-scoring axes and rescores until the plan clears 90.
The rubric weights evidence grounding and dimension coverage at 30 points, concreteness and verifiability at 20, required structure and risk/reversibility at 15 each, and test coverage plus scope discipline at 10 each, summing to 100.
Yes, it reads .harness/config.json for plan.dirs.active, plan.dirs.done, plan.dimensions, and planTypes to match repo conventions. If no config exists, it falls back to a generic template with Summary, Steps, Risks, Test plan, and Standards sections.
Skip it for trivial one-line changes where scoring ceremony costs more than it returns, for one-pass plan reviews where the user does not want iteration, and for emergency incident fixes where stabilization speed matters more than plan polish.
Evidence citations must resolve against the actual working tree. A path:line reference that does not exist is treated as a hard failure of the evidence grounding axis, not a minor deduction, because unverifiable evidence undermines the entire scoring loop.
andreadellacorteResearch codebase patterns and constraints to generate an implementation plan.
ulises-jeremiasGenerate and update project documentation from repository structure and git history.
pem725Analyze user ideas with multi-layer adversarial critique to surface blind spots.
potetoExecute backlog tasks via scoped plans, worktrees, tests, and commits.