Paper Init

Build the local context bundle a later session needs to implement a research paper: the paper itself, its figures and references, what it specifies, what it leaves open, and its data. Context only — no implementation code, no stack setup. Everything the implementer needs must end up on disk; nothing may depend on this conversation's context.

The bundle

One paper/ directory at the project root holds everything paper-derived:

• paper/README.md — the bundle index (step 8).
• paper/source/ — pristine originals only: the PDF (supplementary material included) plus the best machine-readable form available (EPUB, HTML, XML, arXiv LaTeX source), exactly as acquired. Derived files (text dumps, extracted markup) never live here.
• paper/source/metadata.json — the provenance record (step 2).
• paper/paper.md — the faithful working transcription (step 3).
• paper/tables.md — overflow tables, only when the transcription takes its escape hatch (step 3).
• paper/figures/ + paper/figures/README.md — every figure and its ledger (step 2).
• paper/references.md — every reference, annotated (step 3).
• paper/cited/ — pristine sources of the works the paper, or a fetched work in turn, defers load-bearing content to (step 3), when any were fetched. Nothing derived lives here.
• paper/spec.md — what the paper actually specifies (step 3), including its original reported targets with stable identifiers.
• paper/acceptance.md — for empirical papers: selected scope, source-linked requirements, experiment protocols, acceptance criteria, and the evidence a later implementation must produce. Prepared in step 3, reconciled in step 8.
• paper/open-questions.md — what the paper leaves open (step 3).
• paper/dataset.md — dataset provenance and re-acquisition, when the paper uses datasets (step 6).
• paper/reference-code/ — vendored existing code, when it exists (step 4).
• data/ — the datasets themselves, gitignored, when any were acquired (step 6).

Keep generated workflow framing tool-agnostic: "a fresh session", "the implementing agent", without /optimus:* commands. Preserve product/model names that occur in the paper or its evidence; faithful transcription takes precedence. The final chat message may name /optimus:* commands.

1. Resolve the paper

The invocation argument is a URL, a local PDF path, a DOI, or an arXiv id; if none was given, ask for one. Resolve DOIs and arXiv ids to the source of record. If the paper is inaccessible (paywall, dead link), say so plainly and either stop or proceed from a file the user supplies.

Use any supplied scope, resource budget, and reference-code restrictions; retain earlier authorized decisions without asking again. An unspecified budget does not authorize paid compute or services, or large downloads.

When git rev-parse --is-inside-work-tree returns true, resolve its --show-toplevel and proceed in that working tree, including linked worktrees. Otherwise read $CLAUDE_PLUGIN_ROOT/skills/init/references/multi-repo-detection.md and apply it: the bundle goes inside the target repo, not above it. When it finds one or more child repos, ask which one the paper work targets (confirm a lone one instead of suggesting cd) before writing anything — bundle, .gitignore, and README block all land there; when it finds none, work in the current directory.

If paper/ or a papers/*/ directory already holds a bundle, read references/re-running.md and apply it before the first write.

2. Acquire sources

Download into paper/source/, redundantly: the PDF whenever one exists (an HTML-only paper's publisher full text is the primary source), any supplementary material, plus the cleanest structured full text the publisher offers — the transcription cross-checks formats against each other. For arXiv papers, also pull the e-print source bundle (https://arxiv.org/e-print/<id>) when offered: the LaTeX source makes math transcription near-mechanical and ships figures at native resolution. Pull figure rasters into paper/figures/ from whichever source has the best resolution (PDF-embedded usually beats web-served); keep native formats, never re-encode, and write paper/figures/README.md as they land — one line per figure (file, dimensions, caption from the paper text), known defects (duplicates, missing diagrams) at the top. Installing transient fetch or extraction tooling along the way (a PDF library, gdown, pandoc) is fine — that is not the project stack — but install it isolated (pipx, a scratch venv, pip install --target into a temp dir), never into the project's own environment.

Acquire public author clarifications, errata, and addenda that affect the method or results, recording their source and date. Do not contact authors without explicit authorization. Distinguish the paper version from later clarifications and keep conflicts visible until resolved.

metadata.json records at minimum: title, authors, venue, published, doi, url, license, downloaded (date), code_available (with the paper's own availability sentence when it states one), dataset_referenced (name, URL, whether the paper redistributes it — full provenance and re-acquisition live in paper/dataset.md), source_revision (version or publication revision), scope_decisions (every scope decision, supplied at invocation or taken in step 5: decision, rationale, affected target IDs, scope revision), and local_files — every acquired source file mapped to its role, SHA-256, and exact acquisition record (URL or command, and date; for a file the user supplied, the path it came from). Do not hash the metadata file into itself. Include clarifications and supplementary sources. Add any further bibliographic fields the source offers. The test: a fresh clone can re-acquire every publicly fetchable file from this record alone.

3. Working forms

Extract a short paper inline. Otherwise delegate substantial, independent source sections to readers launched as Agent tool calls in a single message, each given explicit sources, required provenance, and absolute paths; reconcile conflicts and coverage yourself before writing the final bundle.

• paper/paper.md — a complete transcription, not a summary: mirrored section headings, math in LaTeX, figures as local relative links with their captions, tables inline (escape hatch: a separate tables.md when tables are numerous or large, linked both ways). Open with a provenance header naming the source of record. Its length is the paper's own.
• paper/spec.md — only what the sources state, each fact tagged with its source and section: data and preprocessing, architecture, training procedure (including stated hardware, time, scale, and seed/repeat policy), evaluation and metric definition, baselines, ablations, and reported results. Architecture/training/eval tables carry a "defined enough to implement?" column. Anything inferred, chosen, or assumed is labeled as ours or moves to open-questions.md — never present our choices as the paper's. End with a short "reported targets" section: externally meaningful results (numbers, and figures where the claim is qualitative), with metric, data, protocol, and spread the paper reports, plus the cheaper checks the paper states along the way (dataset counts, parameter counts, a loss reached) that let part of the implementation be judged before a full run exists. A target the bundle cannot yet measure — data not on disk, a scorer or protocol undefined — says what is missing. Give targets stable IDs and preserve the original values even when the project selects reduced scope. Link empirical targets to acceptance.md; for a non-empirical paper they are the standalone bar, and a differing selected scope gets a labeled Selected targets section below them. Keep the file under ~200 lines; link detailed method tables when needed, never omit implementation-critical facts just to meet the length guide.
• paper/references.md — every reference the paper cites, annotated: role (dataset, baseline, method), resolved link, fetch priority.
• paper/open-questions.md — everything undefined, ordered by how much it blocks work: missing hyperparameters, ambiguous procedures, figure/table defects (ledgered in figures/README.md — point there, don't duplicate), credibility issues, and a suggested framing for the implementation, tech stack included (drawn from the paper and the project's existing stack if one exists). An open point with a defensible default says so and where it comes from (reference code, a cited work, common practice), so an unattended implementation run proceeds instead of stalling on it. Give questions stable IDs, status (unresolved, assumed, resolved, superseded), affected target IDs, evidence, and the decision's rationale and consequences. Keep resolved choices and deviations as a durable decision history, linking scope decisions to their authoritative record in metadata.json. Mark a finding [verified] only when checked against local files during preparation — never for inference or empirical reproduction. Open with that legend. Settle what the sources can settle now: no verifiable-now TODO leaks into implementation. Keep it under ~150 lines, moving resolved history to linked overflow when necessary.

For a paper with empirical claims, read references/empirical-reproduction.md now and prepare acceptance.md using its contract. For a survey or proof without empirical claims, skip it.

When the paper defers load-bearing content to a citation — an inherited architecture, a borrowed training procedure, a dataset defined there — the bundle's contract covers that content too: fetch each such cited work's sources now, as in step 2, into paper/cited/<slug>/, and record each in a cited_works array in metadata.json (title, identifier, source revision, slug, license, acquired-file hashes, and the acquisition record — URL and date, or the path it came from for a user-supplied file). Pull the specific facts it needs from it into spec.md or open-questions.md, tagged with provenance, and note in references.md why it was fetched — a work the paper itself never cites gets its own entry there. Apply the same test to each fetched work: when it defers content the implementation still needs to a further work, fetch that one too. Sources and targeted extraction only — no bundle per cited work. Expect a handful of works, not a bibliography crawl: the test is dependence, not relevance. An inaccessible cited work stays a visible gap — record what is deferred to it in open-questions.md and report it in the final message so the user can supply a copy.

If producing the bundle took mechanical extraction work a fresh session could not trivially redo (pulling rasters out of a PDF, dumping text from an EPUB), leave one small regenerator script that reproduces those derived artifacts from paper/source/ offline (placement follows project conventions; when the project has none, paper/). Stamp only script-produced files with a do-not-hand-edit header — the model-authored files (paper.md, spec.md, and the rest) stay hand-editable, and a re-run updates them in place. When you leave one, ensure .gitattributes pins the script's outputs — eol=lf for generated text, binary for extracted rasters — so regeneration stays diff-clean on any platform.

4. Reference code

Check the paper's own links and project page, then search for official or third-party implementations, subject to explicit user restrictions. Existing implementations are allowed reference material by default. When access and the user's scope permit, vendor available code into paper/reference-code/ (gitignored — step 7), record its provenance as a reference_code object in metadata.json (upstream URL, exact commit — resolve a tag to its commit when possible — vendor date, license). That tracked file lets a fresh clone re-acquire the code; the vendored directory is gitignored. Add a what-it-reveals summary to references.md (hyperparameters, architecture details, training procedure). Preparation only inspects and vendors it; a later implementation may reuse code within the user's scope and license, with attribution. Record whether it was inspected, unavailable, or excluded by the user's restriction separately from whether it exists. Use code_available: false only when no implementation was found, not for a failed fetch or prohibited inspection; unknown existence stays explicitly unknown. Paper/code conflicts go into open-questions.md, not silently into the paper's specification.

5. Feasibility

When the paper's experiments plausibly demand substantial compute — model training, large-scale simulation or rendering — assess feasibility before step 6 acquires anything big. Most papers have nothing to gate (a survey, a proof, a small-scale study); skip this step for them entirely.

• Draw the requirements from what spec.md recorded (step 3) and from the vendored reference code's own docs, which often state hardware. When neither states them, estimate from what the paper does record — model scale, dataset size, training steps — and label the figures as estimates.
• Cover the selected experiment matrix, including repeats, baselines, ablations, evaluation, storage, wall time, and external services. Record portable requirements and supplied budget constraints in acceptance.md (or open-questions.md when non-empirical), separating stated costs from estimates. Cheap checks and reduced runs are useful separate targets; they do not establish the original full-scale claims.
• Detect the local GPU (model, VRAM) and RAM, and compare against what faithful reproduction needs. Detection tooling is not universal (no nvidia-smi on AMD or Apple Silicon machines): when it cannot answer, ask the user what the machine has — never read a failed detection as "no GPU".
• When the gap or an explicit budget makes the selected targets unreachable, rather than merely slower, ask once with AskUserQuestion: header "Resources", the question naming the limiting factor and the concessions that would close the gap (reduced scale, a dataset subset, quantized or distilled variants, different hardware), options "Continue anyway" / "Reduce scope" / "Pause — line up other hardware first".
• Record each decision in metadata.json's scope_decisions. A reduced scope changes the selected bar in acceptance.md (or a non-empirical spec.md's Selected targets section), leaves the original reported targets intact, gets a linked line in open-questions.md, and may shrink step 6's downloads. "Continue anyway" leaves the targets untouched and notes the resource risk; it does not authorize extra spend. "Pause" stops the spend, not the bundle: step 6 writes re-acquisition steps into dataset.md instead of downloading, open-questions.md records the pause, and the remaining steps finish the context bundle with explicit gaps so a re-run can resume it.
• Never write the hardware inventory itself into the bundle — it stays machine-agnostic; only the decision and its consequences go on disk.

No substantial compute, or no mismatch: no gate, and the final message says at most one line about feasibility.

6. Datasets

Identify every dataset the paper uses. Freely downloadable ones go into data/ now; verify what arrived (file counts, sizes, integrity) against what the source promises, and record the verified numbers. Write paper/dataset.md: provenance, exact re-acquisition commands, the verified counts, license and redistribution terms, and anything deliberately not downloaded. Record dataset revision, splits and selection, raw-to-derived processing (filters, normalization, augmentation), leakage controls, and identity checks where specified. Link protocol gaps and experiment IDs; do not invent preprocessing to fill missing details. Keep it under ~200 lines, using linked overflow for necessary detail. If the paper uses no external datasets, say so in one line of paper/README.md's status and skip dataset.md, data/, and the gitignore pair entirely.

Before a large download (GB-scale or hours of time) not already authorized, confirm with AskUserQuestion — header "Download", question stating size and source, options "Download now" / "Skip — write re-acquisition steps only". Small datasets download without asking. When a download is blocked (auth, license acceptance, a manual form), do not ask — write the exact steps into dataset.md and flag it in the final message.

7. Gitignore and routing

Preserve acquired originals byte-for-byte across Git checkouts. For source/cited files intended for tracking whose acquisition hashes identify exact bytes, add narrowly scoped -text entries to .gitattributes so Git cannot normalize their line endings; exclude model-authored metadata and working forms. Verify with git check-attr text -- <files> that each such file reports unset, and that any staged or committed copies still match their acquisition hashes. Do not change the treatment of unrelated user files. For datasets kept outside Git, the re-acquisition record and integrity checks remain authoritative.

Each .gitignore rule below is independent — apply every one whose condition holds, adding only what is missing:

• Datasets in use: data/* plus !data/README.md.
• Vendored reference code: paper/reference-code/.
• License: when metadata.json's license does not permit redistribution (typical for a paywalled publisher PDF), also paper/source/* and paper/figures/* with !paper/source/metadata.json and !paper/figures/README.md exceptions — a fresh clone re-acquires those from the metadata record, which must therefore stay committed. When the license forbids redistribution or derivatives (an ND clause), flag in the final message that paper.md is a full-length derivative: committing it is the user's call when the repo is or will become public.
• Cited works: the same license test applies to each cited work on its own, whatever the main paper's license — one whose license forbids redistribution adds its own paper/cited/<slug>/ directory, never the whole paper/cited/ (other works may be committable). A fetched work is re-acquirable from the record; a user-supplied one exists only at the path it came from — say so in the final message.

A pre-existing data/ line (directory form, common in ML repos) defeats the !data/README.md exception — git cannot re-include a file under an excluded directory. Narrow that line to data/* (same ignore coverage; exceptions become possible), note the change in the final message, and verify with git check-ignore -q data/README.md: it must exit non-zero, finding nothing to ignore (under -v, a match on the ! line is the exception working, not the file being ignored).

When the paper uses datasets, also write data/README.md: what goes here, the counts when known, license terms, and a pointer to paper/dataset.md. Not a git repo? Skip the .gitignore part and note it in the final message.

If a root README.md exists, maintain one short routing block there pointing at the bundle(s), wrapped in marker comments that identify it as managed:

<!-- paper-context:start -->
## Paper context
...
<!-- paper-context:end -->

On re-run, rewrite only what lies between the markers, never duplicate the block; a ## Paper context heading without markers is the user's own — leave it alone and append a new marked block. If no root README exists, skip — the bundle indexes itself.

8. Reconcile and hand off

Reconcile spec.md, acceptance.md when present, dataset.md, and open-questions.md against the acquired sources, code findings, and verified data. Check every empirical claim has a scope disposition, every selected requirement has source and evidence links (or an explicit gap), and each experiment covers its required method, controls, and protocol. Settle newly answerable questions and preserve unresolved ones; do not manufacture results. Record the acceptance revision and its defining documents in source/metadata.json when present. Write the index, paper/README.md, under ~50 lines: what the bundle is, read order, selected scope, preparation status (complete or complete with gaps), remaining blockers, and evidence status (initially unverified).

For empirical papers, put the empirical reference's fresh-session handoff in the index.

Close with: what the bundle contains and where; cited works — which were fetched and why, which the user supplied, and which remain gaps in open-questions.md (or that none were needed); dataset status (downloaded and verified, skipped, blocked — with the instructions pointer — or none); hardware feasibility (one line when fine, the mismatch and the recorded decision when not); any transient tooling installed; and the suggested tech stack as recorded in open-questions.md — a suggestion only, nothing is installed.

Then: commit the bundle first (e.g. with /optimus:commit, staying in this conversation) so the implementation loop starts from a clean, tracked baseline — the gitignored parts (data/, paper/reference-code/, license-ignored sources) exist only in this checkout, so a worktree or fresh clone re-acquires them from the records or copies them in. Then start a fresh conversation with /optimus:gauntlet, printed as one paste-ready line: the goal names the bundle root with its index as the read-first entry, the bar its acceptance.md and referenced protocols/evidence contract (non-empirical: spec.md's Selected targets section, else its reported targets, plus the unresolved questions). Tell the fresh session to settle acceptance choices and pin that bar before the first critic remit, that passing it means faithful reproduction within its justified criteria rather than outperforming the paper, and to name the selected scope in its result: reduced scope is never reported as reproducing the full paper.

Boundaries

• Never write under docs/specs/ or docs/product/ — /optimus:tdd auto-detects build specs there and /optimus:brainstorm scaffold owns the steering cascade. The transcription mirrors the paper's own headings verbatim, Scenarios included — spec auto-detection reads only those two directories, so no bundle heading can misfire it.
• Never write .claude/.optimus-version (owned by /optimus:init) and never edit .claude/CLAUDE.md (audited and reconciled by init under its preservation rules).