paperthin

Land handed-over material of any shape (a note, a link, a dump, a correction) in the document of an existing knowledge base that owns it: read the base's own routing documents first, edit in each document's own grammar, hash files before writing, and report what cannot be placed instead of guessing. User-invoked. Not for researching the subject or designing a new structure.

Keep a repository's README for the person deciding whether and how to use it, its AGENTS.md (or a CLAUDE.md read through it) for the agent working inside it, and the manifests hosts read first (package.json, plugin and marketplace files, agents/openai.yaml) on the repository's one name: audit them with the bundled script, then move each line to the file whose reader needs it and fix each disagreeing manifest. Use when writing, rewriting or reviewing a README or an AGENTS.md, when an agent-only command shows up in a README or marketing copy in an agent guide, when deciding which file a section belongs in, or after a rename. Not for contracts, ledgers, changelogs or release notes.

Keep one paperthin install current and loaded once: move an old skills.sh install onto the Claude Code or Codex plugin, clear the session-start notice older releases wired, set up the daily update job where skills.sh stays, and bring that install to the full current catalog, all shown and confirmed first.

Consolidate duplicate files or directories into one canonical home after approval, held as an identity by this machine's filesystem: hardlink, symlink, junction, or bind mount, with bytes, ACLs, owner and mode, named streams, xattrs and kept timestamps preserved. Use when consumers need the same bytes at several paths and textual references cannot serve them, when asked to hardlink, symlink, junction or bind-mount duplicate copies, or to dedupe identical files or trees on disk without breaking their readers.

Audit and consolidate a fact that's scattered across places into one canonical source, after approval, and replace the rest with references. Use when asked to find duplication, check consistency, locate the source of truth, deduplicate, consolidate, unify, or establish SSOT across artifacts or platforms. Starts read-only, reports the map and plan, then mutates only after explicit approval.

Feed what a finished cycle taught back through the sources that produced it: measure the lesson, question the requirements it touches, land it in one owner, refresh that owner to a clean v0, cut its padding and realign its listings, each pass only when it applies. Use after a correction, a request the user had to repeat, a failed or disappointing cycle, or a cleanup pass over a repository's docs, rules or skills, when the lesson has to change the sources rather than sit in a memo.

Rebuild the human's lost context on a project from live state, in plain language: what needs them, what changed, what new words mean. Use when the human returns after a gap, says they can't follow the project anymore, asks what happened or what a term means, or before deciding what to do next when their mental model is stale. Read-only; it briefs, it does not act.

Read the live cycle state and return the single highest-leverage next best action, not a menu. Use when a project is between phases, the author asks what to do next, too many valid threads are open, or the work needs re-entry into frame, build, drive, re0-memo, hate, re0-work, or ship.

The one command to remember: type /paperthin, then your dream. It hands the request to re0-workflow, which runs the rest.

Run repeated cycles of build, QA, re0-memo and re0-work while preserving learning and letting code die. Use for long agentic projects where progress must be measured by quality-cleared templates, reusable modules, and eliminated anti-patterns rather than hours spent or features accumulated.

Turn a finished, failed, or disappointing work cycle into portable lessons, anti-patterns, quality gates, and next-cycle vocabulary, and measure how far the work stands from its objective. Use after a build, QA pass, demo, user complaint or correction, a request the user had to repeat, a user asking why something was not done, a lesson whose own fix caused a new failure, or an abandoned attempt, and always before a restart from v0, when the useful output is what the next pass must learn rather than the code itself.

Open a paperthin iteration's casebook before re0-loop's first turn, seeded with real content the moment the folder exists, never an empty directory. User-invoked, paperthin-only: assumes the full skill package installed. Run when opening a new build cycle in this repo.

Learn one skill of the suite through three graded levels whose completion rests on the learner's own submission, never on an invocation trace.

Watch a long-running agent job for stalls and alert by default, with recovery only as a proposal a human approves.

Restart a project or artifact from v0 while preserving only proven lessons, contracts, gates, vocabulary, real-surface tests, and negative corpus. Use when the foundation is wrong, accumulated code is misleading progress, or a new pass should learn from the old one without inheriting its accidental architecture.

Read one stated intent, pick and order the installed skills that serve it, show that order, then run the steps the session may run and stop where a person must act. Use when someone states a goal but does not know which skills serve it or in what order, asks which skills to use, or wants a job that needs several skills run end to end. Typing /paperthin lands here.

When data lands with a thin ask (files, a repo, a paste, a chat scrollback), read it and propose what the user is trying to do with it for them to confirm. Use the moment material arrives without a fully-formed ask, or when the ask reads thinner than the data suggests.

Carve guardrail-adjacent items out of scope with safe alternatives before risk-adjacent work starts, then run the safe remainder at full strength in a fresh subagent that only ever sees the carved prompt, never the risky input. Use when a task includes stealth, scraping, privacy, IP, policy, licensing, security, or other safety-adjacent material that could be silently dropped, over-elaborated, or needlessly diluted. Fires on the impulse, not only the topic: the moment you notice yourself about to hedge, soften, silently skip, or brace for a refusal, carve before you execute.

When every option on the table is a variation of one route or waits on the same sign-in, approval, person or dependency, list the options that frame already gives, then map working mechanisms from five or more distant domains onto the problem and report the ones the first list lacked, each with where it breaks. Use when work is stuck inside one frame, when the user asks how another field or industry would solve it, or asks for an analogy that actually transfers. Not for re-reading a decision at a larger scale (macrothink), listing options when no single frame is stuck, or checking claims (factchk).

Cut an artifact that has accreted into bloat, prose or code, to its load-bearing density, meaning and behavior preserved. In prose: padding, over-qualification, fused sentences, walls of enumeration, adjacent restatement. In code, behind green tests: comments that restate the code, commented-out code, vague TODOs, dead code, debug leftovers, guards on guaranteed values, pass-through wrappers, single-use helpers, and nesting a guard clause flattens. Use when the user names an artifact, section, file list, or diff that is correct and current but verbose, patched-over, or slop-filled and wants it tight without a full rewrite; when they ask to clean up, deslop, or remove AI slop from code or recent changes; or when a pipeline the user invoked passes that scope on. Acts only inside that scope.

Remove the em and en dashes an agent drops into running prose where a comma, colon, parentheses or a sentence break belongs, from a user-owned scope, keeping hyphens, ranges and dashes that separate a label from its text.

Review the middle dots (U+00B7) in a Korean passage, field, section, file or folder the writer names, and propose a per-dot replacement only where a dot strings together an open list, keeping every legitimate or protected dot and leaving the text unedited.

Find one person's or one organization's data leaked into a shared repository, folder or artifact, using needles built from that owner's own records (names in every script, handles, emails, phone numbers, addresses, account and client ids), replace each hit with a role or placeholder that keeps the artifact working, and scan git history read-only for the commits that still carry them. Use when a public or shared repository, a template, a fixture set, docs or a deliverable may hold an operator's, a client's or a teammate's personal or business data, when someone asks to de-identify, anonymize, scrub or depersonalize an artifact, before publishing or open-sourcing a repository, or when a person's data must leave a codebase every operator runs.

Find the mocks, stubs, fakes, hidden fallbacks and placeholder implementations standing on a real code path outside the tests, where nothing real runs and a success report is false, and propose the real dependency or an honest failure for each. Use when a change or a named file, directory or diff may ship a test double or a swallowed error into production, when a catch or except returns an empty list, None, a default or true, when a guard lets a request through because its key, config or input is missing, when a function raises NotImplementedError or returns a TODO value on a live path, when a feature reports success but nothing seems to happen, or before calling work done that touched error handling, integrations or service clients.

Make durable artifacts portable by replacing incidental stack, vendor, model, CLI, path, quota, or harness nouns with the mechanism they mean, while keeping provenance, runbooks, and tool-subject claims concrete.

Question a set of requirements or running commitments (a plan, spec, backlog, process, rule set, or one person's goals, projects and side projects) in Musk's fixed order: make your requirements less dumb, delete the part or process, simplify or optimize, accelerate cycle time, automate. Name the goal first, give every item a named source, what it buys, what depends on it and what re-adding it costs, propose removals and deferrals, stop for approval, and never touch an externally mandated constraint. Use when a plan is about to be built as written, or when goals, commitments or side projects compete for the same time.

Verify reality-grounded claims against external sources in both directions before they ship: could the 'absurd' be real, could the 'obvious' be false or long-established? Use whenever an artifact, or the sentence you are about to write, asserts something as plausible, realistic, absurd, novel, or impossible from intuition rather than a checked source; before relying on a factual claim in prose, a design rationale, a research claim, or a plan. Fires on metacognitive doubt: when you can't actually know, verify instead of trusting the feeling. Scans read-only, then fixes the clear errors or flags the judgment calls; leaves deliberate fiction alone.

Attack a plan, design, or argument like you want it to fail before you commit real effort; return the single load-bearing objection and the cheapest experiment that would prove it matters, not a checklist. User-invoked on purpose: a hate-it reflex always in the agent's reach would bias it toward demolition. Weighs many failure axes but owns the synthesis to one root.

User-invoked read-only pass for checking whether the current direction is tunnel-visioned: strip the session's bait, fan out 2 to 5 same-model fresh reads, and report divergence first without treating convergence as proof; or, when asked, re-read one decision at a larger scale and return linked expansions, a no-change, an abstain or a handoff.

Audit any eval, metric, experiment, or benchmark for leakage: does external ground-truth enter independently, or are the model, scorer, and designer just confirming a result no outside truth ever produced? Use before trusting any 'how we'll know it worked', including an A/B, a holdout, a score, a validation, and whenever a result feels too clean or self-confirming. Walks an 8-pattern leakage taxonomy and returns only the patterns that fire, each with an independence fix. Read-only.

Size a task's run before spending it: the cheapest sufficient capability tier (fast, standard, frontier) and the reasoning effort within it, on a neutral scale that binds to whatever levels the model exposes. Use when work seems over- or under-powered, costly, ambiguous, or high-risk, or asks which model class and how much thinking is enough.

Turn material submitted for a decision into a traceable, checkable instrument whose verdict is computed from the decider's own answers, never recommending an outcome.

Rewrite a finished commit's message into a clean, handoff-ready form in your own log style, and place a late commit in time order on a linear history, so `git log` alone tells the story. User-invoked: run it once you've decided to commit.

Review and land an external contribution the way this suite does: gate it against the thesis, land it with the author's credit intact, complete a new skill rather than merging it raw, then approve, credit, and explain before closing.

Move items only, reword nothing; realign a listing whose order has gone arbitrary (a list, table, catalog, sections, an enum) into a logical sequence under one stated principle. Nothing is added or removed. Use when the user names a listing, or a scope holding one, whose order no longer helps a reader follow the set, or when a pipeline the user invoked passes that scope on. Acts only inside that scope.

Walk a pending change through this repo's shipping and releasing checklist end to end, then tag and publish once confirmed. User-invoked: run it when you've decided to ship.

Republish a GitHub repository as one signed commit: squash its whole history into one commit with the same tree, then delete the remote and recreate it under the same name and settings, because a force push leaves the old commits reachable by hash, in forks and in pull requests. User-invoked, and the deletion cannot be undone.

Fix code in a named change or file list so it matches the conventions its own repository already keeps: run the project's configured formatter and linter in fix mode through the project's own toolchain, then fix by hand what no tool decides (naming, import order, file layout, error style) toward the dominant pattern in neighbouring files, behavior preserved, and ask with counts where the repository keeps two competing conventions. Use for 'make this match the codebase', formatter or linter setup, naming consistency, or a style pass on a change under review. Acts only inside the change or files the user names.

Refresh an existing artifact into the current best v0. Use when the user asks to clean up, sync up, dedupe, de-noise, smooth, rewrite, or update an artifact after iteration; when nearby artifacts may have drifted; or when changes in one place should be reflected across related artifacts while keeping the result minimal.

Verify the model's understanding of a user's instruction before spending non-trivial work. Use when a request is long, bundled, high-stakes, hard to undo, or has ambiguous scope or referents such as this, that, it, the other one, whatever is cleaner, or whichever order makes sense. Restate internally, cross-check against available context, proceed silently when resolved, and surface only a genuine surviving fork.