eric-tramel/cross-harness-skills/crystallize
Claude Code and agent skill by eric-tramel/cross-harness-skills
- community
Checking that the SKILL.md file is still in the repository.
Turn a rough feature, bug, refactor, architecture, documentation, or operations idea into a ready-to-implement local plan file. Use when the user asks to crystallize, shape, refine, plan, spec, or de-risk ambiguous work before implementation; the workflow researches the codebase, public prior art, and relevant history, debates the proposed approach with subagents, and writes a final plan with no open questions.
Install
npx vskill@latest install eric-tramel/cross-harness-skills/crystallizeSKILL.md
Crystallize
Overview
Use this skill to convert rough input into an implementation plan that another agent can execute without needing the original discussion. The output is a single ignored local Markdown file under the gitignored plans/ directory.
Do not implement the plan while crystallizing. Inspect, research, and reason; only write the plan artifact.
Plan Artifact
Create plans/ if it does not exist:
Choose a unique filename:
- Derive a lowercase hyphenated slug from the rough input.
- Prefix it with a UTC timestamp from
date -u +%Y%m%d-%H%M%S. - Use
plans/<timestamp>-<slug>.md. - If the file already exists, append
-2,-3, etc.
Keep plans/ and every plan file ignored and untracked. Do not add them to git, do not create a tracked placeholder, and do not include plan files in commits.
Research Wave
Launch parallel subagents before drafting the plan. Give each subagent the raw user input, any linked issue or document text, and explicit permission to inspect the workspace. Do not include your intended solution.
Use these research roles unless a role is plainly irrelevant:
CodebaseResearch: identify affected packages, modules, APIs, tests, docs, and local patterns. Return concrete file paths and constraints.HistoryResearch: use available session-history or agent-history tools when prior or active agent context may matter. Search past sessions by keywords, inspect relevant sessions, and return only actionable history.PublicPriorArtResearch: research public implementations, official docs, standards, libraries, and comparable approaches. Prefer primary sources and include links when external sources influence the plan.RiskResearch: identify correctness, security, performance, migration, compatibility, rollout, and testing risks.
Subagent prompt shape:
Wait for every required research subagent to finish. If a subagent fails, retry once with a narrower prompt. If a role is still unavailable, perform that research yourself before drafting.
Draft Plan
Synthesize the research into a concrete plan. Resolve ambiguities yourself using evidence and repository conventions. Do not ask the user open-ended questions unless the request would require an external decision that cannot be inferred safely; if that happens, stop before writing the final plan.
The draft must include implementation steps, affected files, validation, risks, and acceptance criteria. It must be specific enough that another agent can start from the plan file alone.
Debate Wave
Before finalizing, launch a second parallel wave of subagents to argue against the draft plan. Provide the draft and the research summary, not private reasoning.
Use these debate roles:
PlanSkeptic: find bugs, missing requirements, invalid assumptions, and vague steps.PlanSimplifier: force the minimal viable approach and remove unnecessary generalization.PlanIdiomaticReviewer: check language, framework, library, and repo idioms.PlanScopeGuard: reject scope creep and unrelated cleanup.PlanValidationReviewer: verify the validation strategy would prove the change works.
Debate prompt shape:
Resolve every substantive objection. If debate agents disagree, decide the best course using evidence, repo conventions, and the user's original objective. Prefer the PlanScopeGuard when other roles request broader behavior that crosses the task boundary.
Run targeted follow-up only with the debate subagents whose objections required interpretation or whose facet changed after revision. Do not start a whole new debate wave unless the draft was fundamentally replaced.
Final Plan Requirements
The final plan must have no open questions, no TBD, no unresolved alternatives, and no "ask the user" steps. Every uncertainty must be converted into a decision, assumption, or explicit non-goal.
Write the plan file with this structure:
Include exact file paths where possible. Include commands with expected outcomes. For stack-facing changes, include the repository's normal integration or smoke validation. For public prior art that materially influenced the plan, include source links in Context.
Before finishing, run:
The expected result is that root .gitignore ignores the plan file through the plans/ rule. In the final response, report the plan path and state that it was intentionally left ignored and untracked.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingplugins/cross-harness-skills/skills/crystallize/SKILL.md
Popularity
- Repository stars
- 2
- 7-day trend score
- 0
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/eric-tramel/cross-harness-skills/crystallize)