Arch Skill

Answer or rewrite user-facing responses in ELI10/ELI16 maximum-readability style: reduce reader working-memory load, lead with the point, preserve exact technical truth, define load-bearing jargon, explain mechanisms plainly, put meaning before proof, avoid baby talk and fake memory, and use scan markers or tables only when they clarify. Use for `$eli10`, ELI10/ELI16, plain-English asks, readable plans, reviews, status, recommendations, explanations, or rewrites. This is a response style, not the task owner.

Explicitly selected small-feature start/plan/finish lifecycle with one compact plan and local implementation. Use when the user chooses $lilarch or binding task instructions require it. A small feature or improvement alone does not select this lifecycle.

Upload local artifact files or directories (HTML reports, screenshots, analysis bundles, any static files) to the team's Cloudflare R2 share bucket and return a public unguessable https://share.fun.country/<slug>/... URL. Use when the user says 'upload this to cloudflare somewhere I can share', wants a link to drop in Slack, or wants a local HTML report viewable by teammates. Requires a secret env file at ~/.config/cf-share/env (see references/setup.md). Not for deploying apps or product content (ps-content CDN), not for claude.ai Artifacts, and not for material that must stay private.

Explicitly selected ordered process with a clean worker and critic per step and exact-worker repair. Use when the user chooses $stepwise or binding task instructions require it. A numbered task list or ordinary multi-step work does not select worker orchestration.

Explicitly selected delete-first documentation audit and retirement loop grounded in current code. Use when the user chooses $arch-docs or binding task instructions require it. Ordinary README edits, documentation cleanup, navigation work, and applying existing audit findings do not select this lifecycle or authorize historical-document deletion.

Explicitly selected multi-plan execution lifecycle: decompose an authorized goal into approved ordered arch-step sub-plans, then plan, implement, and review them. Use when the user chooses $arch-epic or binding task instructions require it. Filing or decomposing GitHub issues, mentioning an epic, or asking for status does not select execution.

Read-only flow-status and next-step router for arch-step, miniarch-step, arch-mini-plan, and lilarch docs. Use when the user asks 'what's next?', wants a checklist, or wants the single best next move without running the workflow. Not for doing the planning or implementation work itself.

Explicitly selected full architecture lifecycle around one canonical plan: planning, implementation, review, and supported automatic commands. Use when the user chooses $arch-step, an explicit arch-step command, or a selected parent requires it. Ordinary planning, implementation, and status requests do not select this lifecycle.

Required operating contract for Codex CLI browser work and direct BrowserOS MCP use: protect the user's foreground focus on a shared machine, continually verify the working profile/window/page among many open profiles, reuse existing windows, clean up task-owned pages, and recover the same session safely. Work through viable background methods before necessary brief foreground use; no separate focus approval is required. Codex CLI uses BrowserOS only. Use alongside narrower site skills such as $chatgpt-web. Not for BrowserOS installation/vendor development or generic non-browser connectors.

Explicitly selected analyze/fix/review lifecycle around one canonical bug document. Use when the user chooses $bugs-flow or binding task instructions require it. Symptoms, logs, crashes, Sentry issues, failed tests, and ordinary bug investigations do not select this lifecycle.

Explicitly selected worker orchestration for an approved plan or outcome: the parent directs, integrates, and reviews while delegated workers implement. Use when the user chooses $conductor or binding task instructions require it. An ordinary request to finish work or implement a plan does not select a worker fleet.

Run the standalone goal-seeking loop with a controller doc and append-only worklog. Use for open-ended optimization, investigation, or metric-improvement work where the goal is clear but the path is not. Not for fixed-scope architecture plans, small feature delivery, or ordinary bug fixes.

Explicitly selected long-lived companion that resolves claimed blockers against existing user authority. Use when the user chooses $unblocker or a deliberately selected parent workflow requires it. A stalled task or approval question alone does not select this companion; it cannot create permission for production actions or scope expansion.

Explicitly selected repo-wide audit and repair loop: map the codebase, rank consequential defects, fix within current contracts, and review each change. Use when the user chooses $audit-loop or binding task instructions require it. An ordinary audit, known bug, or request to fix a few findings does not select this persistent lifecycle.

Explicitly selected readiness audit of an existing planning artifact, or implementation audit of code against that plan. Use when the user chooses $plan-audit or a selected parent requires its review doctrine. Ordinary plan feedback, code review, and applying existing audit findings do not select this workflow.

Query logged-in ChatGPT through BrowserOS after applying $browseros and $prompt-authoring. Use for explicit web consultations, data questions, pushed PR reviews, attachments, or exact conversation continuation. Requires @BigQuery for data and @GitHub for GitHub/repo questions, with actual source access; missing or broken access means switch Pro accounts or stop and tell the user, never accept guesses. Defaults to GPT-6 Astra with literal Pro and Extended thinking in Chat, never Extra High, xhigh, Ultra, or Thinking. Discover the already-open Work and Pro profiles, note and use a working account, and continually verify profile/window/page. Missing Pro probably means a temporary limit; try another account in the same-named project with context carried over. Uses one tab, serial prompts, and a heartbeat during waits. Pause blocked consultation when no suitable account works; continue independent work. Not for API work, generic browser automation, automated login, or scripts.

Explicit-invocation epic loop, fired only by name or direct command; never self-select it. Works an epic or milestone most-important-first through issue-to-pr to merge-ready PRs. Owns the live queue, persistent goal, unblocker, and shared GPT-6 Astra Pro planning, meaningful batch checkpoints, major-blocker consultations, and final stack review. Astra and Fable coordinators use delegated-implementation: workers code, test, and repair; the parent owns direct review of every deliverable and all skill authorship. Preserve accepted scope and honest receipts; continue independent work around blockers. Never merge or release. Not for status reads, decomposition via arch-epic, or open-ended optimization.

Explicit-invocation issue delivery, fired by name or by user-invoked epic-to-prs; never self-select it. Takes a GitHub issue to a merge-ready PR with a plan on disk, startup-pragmatism, GPT-6 Astra Pro planning/final review, and PR authoring/follow-through. Astra and Fable coordinators use delegated-implementation: workers code, test, and repair; the parent owns decisions, every deliverable's direct review, and all skill authorship. Related issues can share Pro coverage. Preserve accepted scope and review receipts; never merge or release. Not for investigation-only asks, standalone planning, or work without a GitHub issue.

Publish this skills repo across Amir's machine cluster by committing and pushing the current local work, installing locally, then SSHing to the known hosts, skipping the current machine, pulling the same branch from ~/workspace/<repo-name>, and running the local install. Use when Amir asks to distribute, sync, or publish skill changes to his machines. Not for generic deployments, CI release work, or remote edits outside this repo.

Explicitly selected repo-wide comment-hardening loop: map explanatory gaps and improve high-value code comments without changing behavior. Use when the user chooses $comment-loop or binding task instructions require it. Ordinary documentation edits or local comment changes do not select this loop.

Reclaim substantial disk space on Amir's developer Macs by starting with known worktree, build, cache, and log locations and reusing local cleanup inventories. Use for a full disk, accumulated agent worktrees, routine developer cleanup, or a target such as 1 TB free. Protects canonical checkouts, active work, Git commits, and unique local files; verifies actual free space. Use codex-cleanup for a task confined to Codex SQLite/session maintenance, and arch-docs for stale documentation.

Use fal.ai MCP tools or direct fal SDK/API calls to discover or recommend models, inspect schemas and pricing, upload files, run/submit/stream inference jobs, poll results, and handle outputs, errors, cost, and retention. Use for fal.ai tasks such as background removal, image/video/audio/3D/vision generation or editing, and schema-aware request building. Prefer an installed fal.ai MCP server when available and fall back to fal_client or raw HTTP. Do not use for non-fal providers, generic media editing that will not call fal, account administration, or installing/configuring the fal MCP server.

Run reusable higher-level Herdr operations that compose multiple session, workspace, pane, and agent steps. Use when the user asks to migrate, copy, transfer, or reorganize a workspace between Herdr sessions while preserving its directory, layout, and exact resumable agent continuity when available. Not for ordinary one-session navigation, splitting panes, creating workspaces, or raw CLI control; use $herdr for those primitives.

Write and publish high-quality GitHub pull requests from real repo changes. Use when the user asks to create, open, author, update, or publish a PR, or to turn the current branch/diff into a GitHub PR body/title. This owns PR authoring plus GitHub publication; it is not for code review, release notes, changelogs, or merely printing a suggested PR description.

Historical local Codex, Claude Code, Pi, or Prime Agent session evidence: retrieve prior prompts, goals, corrections, tool calls, child-agent briefs, and timelines. Supports natural-language history requests with bundled JSONL/SQLite helpers. Not for live agent status, streaming output, or session control; use the runtime's live interface for those. Defaults to the current project/runtime and last 24 hours.

Monitor and keep alive a long codex goal-mode session that is already running in a tmux pane, until codex's own goal genuinely completes. Watch the pane for usage limits, stalls, or process death; on a real usage limit rotate the aim account (`aim status` then `aim codex use <label>`), then kill, restart, and `codex resume <session-id>` the SAME session, clear the 'Resume paused goal?' menu, and verify it ACTUALLY resumed work on the new account; check on a ~15-min cadence via a background poll, never a tight loop. Use when the user says monitor/babysit/watch/keep-running my codex, keep it going across rate limits, or restart-and-resume it when limited. Not for spawning a worker to do a task (`agent-delegate`), running your own goal-seeking loop (`goal-loop`), a one-off codex review (`codex-review-yolo`), or purging ~/.codex state (`codex-cleanup`).

Purge stale state under ~/.codex that is older than a week so parallel codex CLI/TUI sessions stop locking up on a bloated logs SQLite, an unrotated 4 GB+ TUI log, and months of dated session JSONL. Use when ~/.codex is multi-GB, when 12+ parallel codex instances stall on `database is locked`, when the WAL on logs_2.sqlite has grown past 100 MB, or when you want a routine weekly purge. Not for editing live config (auth.json, config.toml, version.json, installation_id, .codex-global-state.json), not for the codex git repo, not for cleaning unrelated caches outside ~/.codex.

Explicitly selected independent read-only consultation with clean reviewers and exact-reviewer follow-ups. Use when the user chooses $fresh-consult or binding task instructions require it. Ordinary review and requests for a second opinion may use a plain native reviewer without selecting this pass/fail workflow.

Explicitly selected long-lived, read-only advocate checking ongoing decisions against the user's verbatim intent. Use when the user chooses $intent-police or binding task instructions require it. A long task, delegated work, or absent user does not select this companion. Feedback is advisory and cannot authorize scope or production actions.

Explicitly selected trimmed full-architecture lifecycle around one canonical plan. Use when the user chooses $miniarch-step or a selected parent requires it. Ordinary plan writing, feature work, implementation, and status requests do not select this lifecycle.

Run editful Claude, Codex, Cursor Agent, Grok, or Kimi workers as explicit external processes or resumable sessions with exact model/profile resolution and structured receipts. Use when external transport provides a concrete benefit, such as a different provider, a load-bearing exact model/profile, durable exact-session continuation, process isolation, automation, or captured event/final-output artifacts; these are examples, not a closed gate. Route ordinary same-host work directly to the active host's native children when they can satisfy the role. Use exact handles to resume the same external worker. Not for read-only reviews (`fresh-consult`), Codex `-p yolo` reviews, two-model convergence, ordered orchestration, or detached/background delegation.

Explicitly selected one-pass architecture plan with canonical arch blocks. Use when the user chooses $arch-mini-plan or a selected parent requires it. An ordinary short plan does not select this artifact format, and creating the plan does not activate implementation.

Explicitly selected mobile automation audit and repair loop: map journeys and coverage, rank gaps, and verify repairs on the sanctioned device lane. Use when the user chooses $audit-loop-sim or binding task instructions require it. Ordinary QA, a device repro, or a single app bug does not select this loop.

Convert Markdown or document content into FC / Poker Skill branded PDF artifacts using the bundled letterhead CSS, logo, and local renderer. Use for reports, memos, briefs, packets, audits, summaries, exported document tabs, or any ask for an FC-branded PDF / FC letterhead PDF. Not for Drive archival, slide decks, image/video generation, or ordinary chat answers.

Explicitly selected plan implementation lifecycle with an implementation log, proof freshness, and continuing plan-backed review. Use when the user chooses $plan-implement or binding task instructions require it. Asking to implement or finish a plan alone does not select these extra workflow duties.

Explicitly selected evidence-grounded interview that captures an approved Intent Pack before architecture planning. Use when the user chooses $plan-interview or a selected parent requires it. A brief or ambiguous request to build something does not select an interview lifecycle.

Explicit-invocation fleet debrief, fired only when the user invokes it by name ("check my agents") or directly asks what their running or overnight agent sessions actually accomplished and what is ready or stalled; never self-select it to monitor agents inside other work. Verifies every active or recently finished agent lane from artifacts only (commits, changed files, PR state, CI, worktrees, session transcripts), never from agent self-reports, and returns exactly two lists: merge-ready work with receipts (PR, one-line summary, verdicts, CI state, head SHA) and blocked or stalled lanes with the artifact evidence and the one question or action each needs. A claimed "done" without a diff, test run, or PR is reported as claimed-but-unproven. At most one nudge to a stalled worker, then report; this is a debrief, not a daemon, babysitter, or scheduler. Not for single-session archaeology (agent-history) or monitoring inside an actively conducted fleet (conductor owns its own workers).

Explicitly selected dialogue between two model participants that independently reason, critique, and converge on a bounded question. Use when the user chooses $model-consensus or binding task instructions require it. Ordinary analysis, review, and a single second opinion do not select multi-model dialogue.

Write, edit, refactor, or audit prompt-first agent skills so they stay simple by default, self-contained, anti-heuristic, and distinct from nearby peers. Use when a Codex, OpenClaw, OpenAI, Anthropic, or repo-local skill needs stronger intent, trigger boundaries, packaging, references, runtime metadata, or validation before shipping.

Write, edit, refactor, or audit prompts and reusable prompt contracts, including Markdown-backed Codex goal prompt files and paste-sized /goal prompts, so they match the user's intent, evidence needs, constraints, completion rules, and output expectations without becoming brittle or overbuilt. Use for casual prompt-writing asks, system/agent/reviewer prompts, prompt repair, anti-heuristic refactors, and findings-first prompt audits; use $skill-authoring for skill packages.

Write or rewrite reports, status updates, morning and weekly reports, audit summaries, delivery reports, decks, sheet notes, and status answers so a smart, busy expert reader can parse them in one pass: every sentence has a subject and a verb, every thing is named in plain words with its code or number after it, every PR, issue, document, or message is a labeled link, every status carries its cause and next step, numbers arrive with their comparison, the answer comes first, and the report never talks about itself. Use when a reader says an artifact is confusing, hard to parse, too dense, jargon-heavy, or a wall of text, when an artifact must be rewritten so it can be read at a glance without being dumbed down, or as the final writing pass of any skill that publishes a report. Not for public-facing voice or AI-tell gating of marketing prose, HTML report styling, spreadsheet layout, prompt or skill authoring, or deciding what the report should conclude.

Transcribe recorded audio or video speech into a saved plain-text file with the user's OpenAI API access. Use for local M4A, MP3, WAV, MP4, or similar media, especially when the user asks for the latest OpenAI transcription model, long-recording chunking, or explicit parallel processing. Discover the newest eligible model live and adapt to its supported API instead of choosing an older upload model for convenience. Do not use for speech generation, audio editing, summary-only requests, subtitle timing, or live microphone captioning.

Explain the arch skill suite, distinguish the live subskills, and recommend the right one for a user's task. Use when a request asks which arch skill to use, what the difference is between `arch-step`, `miniarch-step`, `arch-docs`, `arch-mini-plan`, `lilarch`, `bugs-flow`, `audit-loop`, `comment-loop`, `audit-loop-sim`, `north-star-investigation`, `arch-flow`, or wants a quick tour of the arch suite. Not for actually running the underlying workflow.

Invoke the local Codex CLI with the exact external `yolo` profile (gpt-6-astra xhigh, fast tier) and captured receipts for an independent review of a substantial artifact or completion state. Use when the user explicitly asks for `codex -p yolo`, an external Codex review, or the profile's exact model/effort/receipt contract. Ordinary same-host Codex reviews should use a clean native child instead. Review-only; requires a concrete artifact or completion target.

Use as a Flutter app-building reference for architecture, code quality, Dart style, state management, lifecycle, performance, rendering, testing, CI, accessibility, localization, security, platform integration, and Flame/game-loop guidance. Use when authoring, reviewing, repairing, or diagnosing Flutter mobile apps or games. Do not use for non-Flutter frontend work, generic visual design, running Flutter tooling, or latest Flutter/Dart facts without checking official docs.

Install the early-stage startup operating frame when invoked: reality-check the current plan, answer, or behavior away from proof-maximizing, receipt-building, scope expansion, and decision avoidance, and toward the smallest useful move, decisions at 50-70% of desired information, and rigor spent only where wrongness is expensive. Use when the user invokes startup pragmatism, says the agent is overbuilding, over-proving, being pedantic, or perfection-maximizing, or wants a plan, recommendation, or in-flight work re-checked through the early-stage lens. Output is subtraction and a forced decision: name what gets cut, what gets decided now at current information, and where rigor is still owed. Not for code-defect review, plan-readiness audit, intent drift policing, or genuinely irreversible high-blast-radius moves (prod data mutation, external sends, money, app-store releases), which keep full rigor.

Write, edit, refactor, or audit `AGENTS.md` files for coding projects so they stay concise, repo-present, plain-English, command-first, scope-aware, and non-inferable. Use when a repo needs a new root or path-local `AGENTS.md`, an `AGENTS.override.md`, a compressed docs index, or a findings-first review of existing agent instructions.

Explicitly selected skeptical review of implementation claims against current code: trace incomplete behavior, competing owners, side doors, and scope drift. Use when the user chooses $cynical-code-review or binding task instructions require it. Ordinary code review or a cold second opinion does not select this specialist workflow.

Explicitly selected case-level population analysis: materialize constituent observations, inspect cohorts and outliers, and reaggregate a bounded statistical question. Use when the user chooses $bottom-up-diagnostic or binding task instructions require it. Ordinary bug diagnosis, logs, and multiple affected users do not select this workflow.

Author, review, or repair Figma files, component libraries, variables, handoff surfaces, and source-backed app/Figma fidelity audits using prompt-first file-craft doctrine. Use for Figma organization, components, tokens, Dev Mode/Code Connect/Make/MCP readiness, duplicate/stale cleanup, screenshot/reference handling, and app/source parity evidence. Do not use for generic visual taste critique, implementing frontend code from Figma, operating the Figma UI/tool itself, or deterministic visual-diff automation.

Write Markdown-backed goal prompt files for ArcStep runs from a user ask plus an optional canonical DOC_PATH, especially `$arch-step auto-plan`, `implement-loop`, `auto-implement`, or `full-auto`. Use when the user wants a durable Codex/Claude goal prompt that references source truth without copying the plan, keeps implementation aligned with plan intent, and catches surface-level completion. Not for executing ArcStep itself or creating harnesses, runners, or controllers.

Build local contact sheet images from existing image files, folders, globs, or attached local image paths. Use when the user asks to make, label, compare, paginate, or adjust a grid/contact sheet of screenshots, renders, before/after shots, visual candidates, or image folders. Defaults to a labeled, dense, near-native, edge-to-edge sheet and opens it in Preview on macOS without questions. Not for generating or editing images, video/GIF extraction, Figma boards, slide/doc layouts, provider APIs, or theme-specific generation workflows.

Explicitly selected review producing a deletion report for low-value code, tests, dependencies, generated artifacts, or documentation. Use when the user chooses $cynical-cruft-removal or binding task instructions require it. Ordinary cleanup requests, docs edits, and applying audit findings do not select this specialist review or authorize deletion.

Explicitly selected code review where systematic coverage is part of the deliverable, with a findings artifact and clean review slices when authorized. Use when the user chooses $exhaustive-code-review or binding task instructions require it. An ordinary code review or completion check does not select this workflow.

Build, restyle, or audit Google Sheets (and Excel) workbooks so a human can read, scan, change and trust them: header-only frozen rows, one line per cell, merges only for group headers, numbers right and text left, consistent number formats, one documented colour key, and every number owned: observations shown as data, assumptions styled as inputs, everything derived as formulas. Use when the user asks to make a spreadsheet readable, consistent, presentable or user-friendly, to fix freeze panes, paragraphs in cells, hardcoded numbers or inconsistent formatting, to apply a house style through the Sheets API or gws CLI, or to audit a sheet's formatting. Not for choosing the analysis or metrics, data cleaning or statistics, chart or dashboard design outside the grid, or HTML/PDF report styling.

Explicit-invocation PR follow-through loop for an already-open GitHub pull request: poll the PR, triage every new review or comment thread with judgment, reply on-thread with accept/decline rationale, push warranted fixes to the same branch, and repeat until the PR is merge-ready.

Audit and score an agent-definition markdown such as `AGENTS.md`, `CLAUDE.md`, `SKILL.md`, `SOUL.md`, `.cursorrules`, or a system prompt using a cold-reader rubric for clarity, consistency, context independence, operating-model fit, and coherent agent-dispatch semantics when present. Use when the user asks to review, score, critique, or audit an agent instruction file. Not for evaluating agent outputs, code quality, or rewriting the file itself.

Rewrite the current branch's branch-span Git commit messages into an informative narrative while preserving patches and commit boundaries. Use when the user wants WIP/vague commits re-authored, branch history clarified from its nearest parent branch, or an active arch plan called out in commit bodies before sharing. Safety-first: infer the parent/base, allow the current branch's own remote ref, block unrelated shared refs, create a backup branch, never push/force-push, and do not squash/split/reorder unless explicitly asked for a separate history-editing flow. Not for PR bodies, changelogs, release notes, code review, or rewriting parent/shared history.

Keep the parent responsible for requirements, architecture, integration, and direct review while its subagents implement, test, and repair. Used by issue-to-pr and epic-to-prs for Astra or Fable coordinators, or when the user asks for this executive/worker split on accepted work. Astra delegates to GPT-5.6 Sol high; Fable delegates to Opus 5. All skill authorship stays with the parent. Owns execution responsibilities, not a delivery lifecycle, agent launcher, read-only review, or conductor's full workflow.

Explicitly selected quantitative investigation loop with a Commander's Intent document, ranked hypotheses, and tests that distinguish them. Use when the user chooses $north-star-investigation or binding task instructions require it. Uncertainty, debugging, or a quantitative question alone does not select this workflow.