avidrucker

Diagnose whether an incoming work-request — a casual task handed to an agent, a ticket, or an ask handed to a teammate — is clear enough to act on, before guessing. Runs a gate (request vs goal vs question vs complaint), counts how many materially-different…

Project management and engineering-discipline skill set distilled from Yegor Bugayenko's XDSD methodology. Meta-orchestrator for 17 sub-skills (yegor-pdd, yegor-spikes, yegor-bdd, yegor-microtasks, yegor-tickets, yegor-architect, yegor-velocity,…

Apply Yegor-style velocity discipline by measuring progress primarily as verified closed tickets per time window, not commits, hours, or lines of code. Use when Codex is reviewing project progress, answering "how is it going?", checking stale work,…

Apply documentation-first NoHelp discipline: reusable project knowledge belongs in repo docs and tracker artifacts, not in chat, memory, or private explanations. Use when Codex is answering project-specific questions, debugging non-obvious behavior,…

Separate architect mode from courier mode: design in writing first, then execute the accepted design without redesigning midstream. Use when Codex is starting fuzzy work, encountering a design question during implementation, reviewing whether a change matches…

Apply ticket-as-conversation discipline by making the issue tracker the primary record for project decisions, assumptions, blockers, status changes, and direction changes. Use when Codex is about to answer a project-specific question, make or record a design…

Apply Yegor-style microtasking by keeping work bounded to small, independently finishable units with a hard 60-minute ceiling and a practical 30-minute default. Use when Codex is estimating, planning, starting implementation, reviewing a large-looking ticket,…

Apply Bug Driven Development by shaping work as a neutral complaint: have X, should have Y, with repro or evidence. Use when Codex is filing, reviewing, rewriting, or closing issues; when a user describes work as a request, suggestion, question, or vague…

Track a body of claims and questions through a human-gated lifecycle — DRAFT → UNVERIFIED → VERIFIED (or CANCELLED), open → answered — where a claim is a validatable statement about code, libraries, or data, and "verified" means a HUMAN judged it, never that…

Audit a software project for WASTE using Sedano, Ralph & Péraire's empirical nine-waste taxonomy (ICSE 2017) — building the wrong feature, mismanaging the backlog, rework, unnecessarily complex solutions, extraneous cognitive load, psychological distress,…

Triage all open issues for the current project and produce a prioritized work plan. Config-driven (.claude/orchestrate.json) — in fleet mode it emits copy-pasteable plain-paragraph assignments per agent (APPLE, BANANA, CHERRY, DRAGONFRUIT, ELDERBERRY, FIG,…

Protocol for recording agent errors into the project's errors store — when to log, what fields to populate, the resolved command, and when to skip. Use whenever a tool call fails, a hook blocks, a claim fails, or any non-trivial error occurs during puzzle…

Pre-close checklist that catches findings that should be filed as tickets before the close commit. Surface it at every close on a substantive puzzle, but run it only with explicit user go-ahead — it is a token-heavy pass, not an auto-run step.

Track wall-clock time per puzzle/ticket with dual H/C estimates and a logged row so forecasts can be calibrated over time. Use when picking up or closing a puzzle, starting work on a ticket, or logging velocity. Pairs with the yegor-* family.

Guide an agent through writing, filing, and closing a TIL (Today I Learned) entry in docs/learnings/. Use when the user says "write a TIL", "write up what you learned", "add to learnings", or requests a session retrospective.

Rank the open issues/puzzles by ICE score (Impact×Confidence×Ease) then severity, and render them with severity-color emoji, splitting out blocked and iceboxed work. Use ONLY when the user explicitly asks "what's next", "what should I work on", "triage the…

Guide an agent through writing, filing, and closing a TIL (Today I Learned) entry in docs/learnings/. Use when the user says 'write a TIL', 'write up what you learned', 'add to learnings', or requests a session retrospective.

Re-explain the agent's most recent substantive answer in plain, accessible, jargon-free English. Use when the user types /eli5, says 'explain that simply', 'ELI5 that', 'in plain English', or 're-explain your last answer'. This is a re-render of existing…

Complete GitHub workflow: auth setup, issues, PRs, code review, repo management, and CI. Use when working with GitHub in any way — authenticating, creating/ triaging issues, opening/merging PRs, reviewing code, managing repos, or troubleshooting CI. Each…

CodeMirror 6 + Shiki theme integration for browser-based code editors and playgrounds. Covers per-theme editor backgrounds, dynamic syntax highlighting via compartments, zero-flicker theme switching, and persistent site-wide light/dark mode. Use when building…

Debug Node.js, Python, and general bugs systematically. Covers node inspect, pdb, debugpy, post-mortem debugging, heap/CPU profiling, the 4-phase systematic debugging methodology, and common pitfalls. Use when a test fails, production crashes, behavior is…

Error logging and session learning capture. Covers structured error logging to SQLite (error types, context shapes, de-duplication) and TIL (Today I Learned) document workflow for session retrospectives. Use when a tool call fails, a hook blocks, a claim…

Use when the user types the exact trigger '/fruit-agent-orchestrate' to triage open issues and produce copy-pasteable plain-paragraph work assignments for each fruit agent (APPLE, BANANA, CHERRY, DRAGONFRUIT, ELDERBERRY, FIG, GRAPE, HONEYDEW, INCABERRY,…

GitHub issue lifecycle: quality review before work, human decision facilitation during work, and pre-close checklist after work. Use when reviewing an issue for agent-readiness, walking through a human-required decision ticket, or running the pre-close…

Covers what goes wrong once an autonomous AI agent can act, not just talk — destructive commands on real targets, over-broad permissions, retry spirals, infinite loops, and cheating to hit an exit condition. Use when an agent holds shell/filesystem/DB tools,…

The slow-accruing debts that look free on day one and send a bill later — comprehension debt (shipping code nobody read or can defend), architectural entropy across sessions, the day-one mirage, the collector's fallacy (capture mistaken for progress),…

What goes wrong because of the LLM context window and memory limits — the agent forgets earlier instructions, skims the middle, overflows and evicts constraints, edits blind once the project outgrows the window, and re-bills bloated history every turn. Use…

What goes wrong when a model invents things in its own generated output — nonexistent code/APIs/functions, fabricated package names, made-up citations, and the 90%-correct answer with one invisible invented detail (wrong default, flag, or signature). Use when…

Run a pre-mortem on a project, plan, feature, or design — the pessimistic/critical twin of brainstorming. Systematically surfaces what could go wrong, ranks each risk by likelihood × blast-radius, and converts it into a concrete check or mitigation. Use when…

What goes wrong when you pick or trust a model on hype instead of evidence, and when a shipped model silently degrades — benchmark theater, wrong-axis "beat the frontier" claims, sim-to-real gap, mistaking an orchestration layer for a frontier model,…

How RAG, grounding, and retrieval pipelines fail — the wrong chunk retrieved on keyword-not-meaning matches, the right document misread on a negation or table, the caveat truncated at a chunk boundary, a stale index serving obsolete facts, top-k bloat burying…

Meta-orchestrator for distilling a YouTube talk into study material. Use when the user gives a YouTube URL and wants any of — SRT captions, a clean speaker-labelled transcript, an executive summary, or a timestamp-indexed key-takeaways doc. Routes to…

Read a talk transcript and produce an executive_summary.md (TL;DR + Salient Points + Outline). Use when the user wants a first-read overview of a talk — prose, structured, and skimmable. No script involved; pure prompt-engineering for Claude.

Read a talk transcript (ideally plus the SRT for accurate timestamps) and produce a key_takeaways.md — a flat list of timestamp-anchored emoji-prefixed bullets for jumping back into the video. Use when the user wants a navigation index, not a read-through…

Fetch the best subtitle track for a YouTube URL via yt-dlp. Use when the user gives a YouTube link and asks for "the captions" / "the SRT" / "subtitles" — or as the first step of the talk-distill pipeline. Probes available tracks, prefers human-authored over…

Convert an SRT (typically YouTube auto-captions) into a clean speaker-labelled paragraph transcript. Use when the user has `captions.SRT` and wants `transcript.txt`. Wraps `scripts/srt_to_transcript.py`, which handles rolling-caption dedup, `>>`…

Write idiomatic, correct LCC assembly (`.a` files) for the lccjs toolchain — base ISA only. Encodes the calling-convention contract, register-role discipline, encoding-range gotchas, and branch-condition semantics that the ISA table alone doesn't convey. Use…

Write idiomatic, correct LCC+ assembly (`.ap` files) for the lccplus toolchain — extended ISA with `.lccplus` directive, `rand`/`srand`/`millis`/`nbain`/`clear`/`resetc`/`sleep`/`cursor` extras, and the `lccplus.js` entry point. Use when the user asks to…

Assess the quality of a goal by scoring it on five axes that each run from a SMART pole to a VAPID pole — Specific↔Vague, Measurable↔Amorphous, Actionable↔Pie-in-the-sky, Responsible↔Irrelevant, Time-bound↔Delayed — then classify it and rewrite it sharper.…

Review code for *complecting* — concerns braided together that could stand apart (Rich Hickey, "Simple Made Easy"). Scans a diff, file, or module against a catalog of complex constructs (state⊗time, objects, inheritance, scattered conditionals, direct A→B…