coordinate-external-ag…
Coordinate independently operated external agents through durable handoffs. Use when work crosses hosts, sessions, accounts,…
@gaelic-ghostInstall SkillDevelopment
Skill
/prepare-language-model-dataset
Prepare language-model training, preference, red-team, or evaluation datasets. Use when collecting, filtering, deduplicating, templating, splitting, or auditing examples for provenance, leakage, consent, or sensitive data.
Claude Code
$ npx -y skills add gaelic-ghost/socket --skill prepare-language-model-dataset --agent claude-code
Codex
$ npx skills add gaelic-ghost/socket --skill prepare-language-model-dataset -a codex -y
Cursor
$ npx skills add gaelic-ghost/socket --skill prepare-language-model-dataset -a cursor -y
Gemini CLI
$ npx skills add gaelic-ghost/socket --skill prepare-language-model-dataset -a gemini-cli -y
GitHub Copilot
$ npx skills add gaelic-ghost/socket --skill prepare-language-model-dataset -a github-copilot -y
OpenCode
$ npx skills add gaelic-ghost/socket --skill prepare-language-model-dataset -a opencode -y
Windsurf
$ npx skills add gaelic-ghost/socket --skill prepare-language-model-dataset -a windsurf -y
SKILL.md
prepare-language-model-dataset.SKILL.mdname: prepare-language-model-dataset
description: Prepare language-model training, preference, red-team, or evaluation datasets. Use when collecting, filtering, deduplicating, templating, splitting, or auditing examples for provenance, leakage, consent, or sensitive data.
Prepare Language Model Dataset
Workflow
1. Define the dataset's task, unit of observation, schema, and intended use.
- SFT: ordered messages or prompt/response with explicit loss-mask semantics.
- Preference optimization: prompt/context plus `chosen` and `rejected` responses from the same comparison unit.
- Reward modeling: response or response pair plus label, score, ranking provenance, and annotator agreement.
- Synthetic data: generator model/revision, prompt, sampling settings, filters, and parent-record lineage.
- Adversarial/red-team data: target policy, attack family, authorization, severity, and safe retention boundary.
- Evaluation fixtures: stable case ID, inputs, expected behavior, grader, tags, and strict separation from training.
2. Record every source, revision, retrieval date, license, consent basis, and transformation. 3. Validate parseability and required fields before semantic filtering. 4. Normalize text and chat templates without erasing distinctions the task needs. 5. Remove secrets and unapproved personal or sensitive data before logging or upload. 6. Deduplicate before splitting. Use content hashes plus near-duplicate analysis where paraphrases matter. 7. Split by contamination boundary—document, conversation, author, repository, or time—not merely by row. 8. Search train data for evaluation overlap and record the matching method and threshold. 9. Measure language, length, label, source, safety, and quality distributions for every split. 10. Copy `assets/dataset-card.md` and document exclusions, known biases, and redistribution limits. 11. Freeze a versioned artifact or deterministic build recipe and record checksums.
Acceptance Gates
- Every example is traceable to a source or a documented synthetic generator.
- Train, validation, and test boundaries match the leakage model.
- The tokenizer and chat template used for length checks match the intended model.
- Rejected records and filter counts are retained without retaining removed secrets.
- A human can reconstruct the released split from the documented inputs and transformations.
Resources
- `assets/dataset-card.md`: dataset provenance and quality template.
- `references/dataset-controls.md`: split, leakage, and sensitive-data controls.
Read more
name: prepare-language-model-dataset description: Prepare language-model training, preference, red-team, or evaluation datasets. Use when collecting, filtering, deduplicating, templating, splitting, or auditing examples for provenance, leakage, consent, or sensitive data.
Prepare Language Model Dataset
Workflow
1. Define the dataset's task, unit of observation, schema, and intended use.
- SFT: ordered messages or prompt/response with explicit loss-mask semantics.
- Preference optimization: prompt/context plus `chosen` and `rejected` responses from the same comparison unit.
- Reward modeling: response or response pair plus label, score, ranking provenance, and annotator agreement.
- Synthetic data: generator model/revision, prompt, sampling settings, filters, and parent-record lineage.
- Adversarial/red-team data: target policy, attack family, authorization, severity, and safe retention boundary.
- Evaluation fixtures: stable case ID, inputs, expected behavior, grader, tags, and strict separation from training.
2. Record every source, revision, retrieval date, license, consent basis, and transformation. 3. Validate parseability and required fields before semantic filtering. 4. Normalize text and chat templates without erasing distinctions the task needs. 5. Remove secrets and unapproved personal or sensitive data before logging or upload. 6. Deduplicate before splitting. Use content hashes plus near-duplicate analysis where paraphrases matter. 7. Split by contamination boundary—document, conversation, author, repository, or time—not merely by row. 8. Search train data for evaluation overlap and record the matching method and threshold. 9. Measure language, length, label, source, safety, and quality distributions for every split. 10. Copy `assets/dataset-card.md` and document exclusions, known biases, and redistribution limits. 11. Freeze a versioned artifact or deterministic build recipe and record checksums.
Acceptance Gates
- Every example is traceable to a source or a documented synthetic generator.
- Train, validation, and test boundaries match the leakage model.
- The tokenizer and chat template used for length checks match the intended model.
- Rejected records and filter counts are retained without retaining removed secrets.
- A human can reconstruct the released split from the documented inputs and transformations.
Resources
- `assets/dataset-card.md`: dataset provenance and quality template.
- `references/dataset-controls.md`: split, leakage, and sensitive-data controls.
9200 skills5 MCP
Stars on GitHub
Momentum
- stars in total
- 9
Other skills on socket.
@gaelic-ghostInstall Skillcoordinate-worktrees-a…
Assign worktree, branch, write, validation, integration, and cleanup ownership before parallel repository work. Use when a worker…
@gaelic-ghostInstall Skilldesign-agent-automatio…
Design framework-neutral agent and automation workflows before implementation. Use when choosing between Codex app automations,…
@gaelic-ghostInstall Skilldesign-agent-eval-work…
Design evaluation workflows for agent, skill, prompt, and automation behavior before implementation. Use when choosing eval…
@gaelic-ghostInstall Skilldesign-n8n-agent-workf…
Design safe n8n workflows with deterministic routing, credentials, idempotency, recovery, local-model checks, drafts, and exact…
@gaelic-ghostInstall Skillorchestrate-agent-work
Coordinate bounded worker tasks with a launch envelope, report-back, escalation, and synthesis contract. Use before spawning,…
@gaelic-ghostInstall SkillMore skills from other plugins.
Auto-invoked
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or…
@obraInstall Skillask-matt
Ask which skill or flow fits your situation. A router over the skills in this repo.
@mattpocockInstall SkillAuto-invoked
Questions about prepare-language-model-dataset.
- Which coding agents can use prepare-language-model-dataset?
- prepare-language-model-dataset installs in Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and Windsurf with the skills CLI, using each agent's flag shown on this page. Flowy's auto-invocation, which fires a skill when your prompt matches, runs in Claude Code only.