claude

instruction-eval

Measure whether an LLM instruction changes model behavior.

4 on GitHubby ryzhakarv0.2.0MITupdated 23 days ago
/

Plugin instruction-eval from marketplace my-claude-skills in ryzhakar/claude-skills. Commands read from the repo's manifest — 23 days ago.

What it installs

Skills (11)

  • calibrate-judge

    Measure a candidate judge against reference raters on a blind sample of the retest's open replies, with acceptance set before any rating, and accept or reject the attempt. Use when the user says "calibrate the judge", "write the judge brief", "judge calibrate", "rate the sample", or "calibration rej

  • check-coverage

    Map each demand to the instruction's sentences or to dropped, and each sentence to its demands, into a revision's coverage record, and refuse any imported sentence. "check coverage", "check against the source", "map demands to sentences", "write 3-coverage.yaml", "coverage check failed", "sentence c

  • diagnose

    Read the open failures of a failed revision for one failing demand and write the hypothesis predicting a movement the noise can show, or stop the series as unresolvable. Use when the user says "diagnose", "why did it fail", "write the hypothesis", "what should change next", or "stop the series". Use

  • freeze-series

    Retest `r1` to measure the noise, set each gate check's gap from it with the keep rule and budget, and freeze every part the gate reads by hash. Use when the user says "retest r1", "ask the readers", "collect the replies", "measure the noise", "set the gate", or "freeze the series". Use when a serie

  • judge-replies

    Prepare one judge agent per packet of a blinded revision, launch them, and collect the verdicts whose anchors meet the calibrated acceptance. Use when the user says "judge the replies", "judge r1", "launch the judges", "collect the verdicts", or "re-judge". Use when a revision holds `6-packets/` and

  • keep-revision

    Apply the frozen keep rule at a series' stop and write the kept revision back over the instruction's live path. Use when the user says "keep the revision", "which revision won", "apply the keep rule", "write back the instruction", or "series keep". Use when the series holds `9-stop.yaml`.

  • list-demands

    List every demand the source, or with no source the instruction, makes, one entry each with its boundary, into the series' demand record and check it. "list the demands", "save the demand list", "what does this instruction demand", "write 2-demands.yaml", "demands check failed". A series folder exis

  • measure-revision

    Ask the reader, judge, detect, score, and gate one revision of a frozen series, then hand it to diagnosis or to the keep rule at a stop. Use when the user says "measure the revision", "gate r1", "score r2", "run r3", "did it pass", or "where does the series stand". Use when a series holds `8-frozen.

  • revise

    Make the hypothesis' one change to the live instruction through the writer and record it as the next revision, refusing any second change. Use when the user says "revise the instruction", "apply the hypothesis", "make the change", "record the revision", "save the cut", or "revision new failed". Use

  • start-series

    Create a series folder for one instruction and its sources, and set up the neutral login whose session drives the series. Use when the user says "start a series", "evaluate this instruction", "measure this prompt", "test this digest", "set up instruction-eval", or "new series". Use when a user holds

  • write-tasks

    Write the tasks tempting each demand and `overreach`, their answer keys, conversations, and open and sealed split, with the conditions, reader, and first gate keys they need, and check them. Use when the user says "write the tasks", "write 3-tasks.yaml", "pick the controls", "set up the reader", or

View source on GitHub

More from ryzhakar/claude-skills

All 11 plugins →

Domain-free skills governing how an agent conducts itself while working, independent of orchestration, memory, or any specific engineering domain.

6 skills · hooks

4

dev-cycle

claude

Development cycle orchestration: specify, plan, implement, review, integrate. A spec-capturer that interviews the owner, worktree-isolated implementers, spec and code-quality reviewers, and a review chain enforced by hooks.

1 skill · 4 agents · hooks

4

Cost-tiered development cycle over dev-cycle: an opus prescriber writes prescriptions that leave no decision, a sonnet suborchestrator runs the loop and both reviews, disposable haiku executors follow the prescription literally. No hooks.

2 skills · 3 agents

4

Software engineering discipline skills, agent-agnostic: outermost-test-first signature-first TDD, systematic debugging, bug triage, architecture improvement, and receiving code review. Used ad hoc or by dev-cycle.

5 skills

4

Playwright test lifecycle orchestrator. One skill drives the full loop: plan from live browser exploration, generate accessible .spec.ts files, execute with failure classification, and self-heal broken locators via deterministic ten-tier recovery with confidence-based PR routing.

1 skill · 4 agents

4

Agent delegation framework and multi-agent research orchestration. Decompose work across model tiers, manage parallel swarms, and govern quality.

2 skills

4

Similar plugins

Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.

2 skills · 2 commands · 2 agents · 1 MCP server

f/prompts.chat 177,027

hookify

claude

Easily create custom hooks to prevent unwanted behaviors by analyzing conversation patterns or from explicit instructions. Define rules via simple markdown files.

1 skill · 4 commands · 1 agent · hooks

anthropics/claude-code 155,476

career-ops

claudecodex

Open-source AI job search that runs locally in your AI coding CLI: scan job portals, evaluate listings into a structured A-H report with a 1-5 score, tailor your CV, track applications

5 skills · hooks

career-ops-hq/career-ops 82,253

agentic-flow (external: maintained in ruvnet/agentic-flow, pinned to commit d19737b, plugin 3.4.0): the agentic-flow MCP server pinned to [email protected] (ReasoningBank learning, ecosystem tools prompt_injection_scan, swarm_partition_tasks, sublinear_solve, sublinear_pagerank and ecosystem_status, multi-provider routing, autopilot learning; readonly profile by default, no tool pre-approved), 2 agents, 4 skills, /agentic-flow-status, and a mod guard on agentic_flow_agent with /agentic-flow-guard

4 skills · 1 command · 2 agents · 1 MCP server

ruvnet/ruflo 81,589

last30days

claudecodex

Research any topic across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and 5+ more sources. AI agent scores by upvotes, likes, and real money - not editors.

1 skill

mvanhorn/last30days-skill 67,453

5 skills + 35 modes + Material Passport pipeline. Includes v3.6.7 cross-model audit gate and v3.6.8 generator-evaluator contract.

5 skills · 16 commands · 4 agents · hooks

Imbad0202/academic-research-skills 56,698

Maintain instruction-eval?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins11 plugins