| name | ip-uspto |
| description | Draft, review, and revise USPTO non-provisional utility patent applications (specification, claims, abstract, drawings, formal sections per 37 CFR) through the canonical anvil lifecycle extended with USPTO-specific phases. |
| domain | ip |
| type | skill |
| user-invocable | false |
anvil:ip-uspto — USPTO non-provisional utility patent applications
The ip-uspto skill produces non-provisional utility patent applications targeting filing at the United States Patent and Trademark Office. It extends the canonical anvil lifecycle (draft → review → revise → figures → audit) with four USPTO-specific phases (intake, inventorship, pre-flight, finalize) and is the proving ground for the N parallel critics, one reviser framework primitive.
This skill targets AIA non-provisional utility applications (first-inventor-to-file framing, post–16 March 2013). Provisional applications are handled by the sibling skill anvil:ip-uspto-provisional (claims-optional, enablement-depth-first — issue #433); design patents are out of scope for v0.
Artifact contract
A patent thread is a single patent application authored across one or more revisions. A thread is identified by a slug (e.g., acme-widget, foo-method). Each thread occupies a portfolio directory that contains:
<portfolio>/
<thread>/ Optional thread root with brief and reference material
BRIEF.md Structured inventor brief (intake output, or hand-authored)
refs/ Optional reference material (transcripts, sketches, lab notebooks)
prior-art/ Operator-supplied prior art (PDFs or markdown summaries)
fto-refs/ Operator-supplied third-party references for FTO triage (ip-uspto-fto input; distinct from prior-art/ — FTO targets may postdate priority)
inventorship.md Inventorship matrix (inventorship phase output)
inventorship-evidence/ Optional (ip-uspto-inventorship --evidence); thread-level like the matrix
inventorship_map.json Element/feature → repo-paths map (semi-manual seed; cached; --reseed discards)
evidence.jsonl Append-only git evidence rows (reduction-to-practice citations only)
.anvil.json Optional per-thread overrides (max_iterations, critic set)
<thread>.1/ First drafted version (immutable once written)
spec.tex Specification (LaTeX, using anvil-uspto.cls)
claims.tex Claims block (independent + dependent)
abstract.txt Abstract (≤150 words, plain text)
drawings/ Figure stubs or rendered drawings
fig-1.tex (TikZ flowcharts) or fig-1.svg / fig-1.pdf
drawing-descriptions.md Stub descriptions for human illustrator (default v0)
_outline.json Section-by-section drafting plan (control surface; see "Outline control surface")
_progress.json Phase state for this version
_revision-log.md (revisions only) Maps prior critic findings to changes
<thread>.1.review/ General reviewer sibling (clarity, structure, voice)
<thread>.1.s101/ §101 statutory subject matter critic
<thread>.1.s112/ §112 enablement / written description / definiteness critic
<thread>.1.claims/ Claim breadth + dependency-tree critic
<thread>.1.priorart/ Novelty / §102 / §103 positioning critic
<thread>.1.audit/ Final fact-check (audit phase only; post-convergence)
<thread>.1.preflight/ Pre-flight mechanical compliance scan (post-revise, pre-review)
<thread>.2/ Revised version (after revise consumes ALL critic siblings)
<thread>.2.review/
...
<thread>.{N}/ Terminal version, marked READY then AUDITED then FINALIZED
<thread>.final/ Finalize phase output (assembled submission package)
spec.pdf
drawings.pdf
ads-placeholder.txt
fee-sheet-placeholder.txt
inventorship-attestation.md
_manifest.json
Versioned dirs (<thread>.{N}/) and critic sibling dirs (<thread>.{N}.<tag>/) are immutable once their _progress.json records the phase as done. Revisions are produced as a new version dir, never by editing in place.
State machine
USPTO extends the standard lifecycle with four phases. Per-thread state, derived from on-disk evidence (not flags):
EMPTY → INTAKE_DONE → INVENTORSHIP_DONE → DRAFTED → REVIEWED → REVISED → … → READY → AUDITED → FINALIZED
↑
PRE_FLIGHT_PASSED gates the loop edge
| State | Evidence |
|---|
EMPTY | No <thread>.{N}/ directories exist; brief may or may not exist |
INTAKE_DONE | <thread>/BRIEF.md exists and is structured (has the intake frontmatter keys) |
INVENTORSHIP_DONE | <thread>/inventorship.md exists with at least one named inventor and a per-independent-claim attribution table |
DRAFTED | Latest <thread>.{N}/ exists with spec.tex, claims.tex, abstract.txt, and _progress.json.draft == done; no sibling critic at the same N |
REVIEWED | All configured critic siblings (<thread>.{N}.<tag>/) at the latest N are done |
PRE_FLIGHT_PASSED | <thread>.{N}.preflight/_summary.md records passed: true (or all blockers were waived) |
REVISED | A <thread>.{N+1}/ exists after prior critic siblings + pre-flight at <thread>.{N} |
READY | Aggregate score from critic siblings ≥39/45 AND no critical flag at latest N |
AUDITED | <thread>.{N}.audit/_summary.md records passed: true alongside a READY version |
FINALIZED | <thread>.final/_manifest.json exists with all required submission artifacts referenced |
Thresholds: ≥39/45 advances (legal/customer-facing artifact per anvil's threshold table). Any §101 critical flag OR §112 critical flag short-circuits regardless of total score — block until addressed. Other critic critical flags follow the same short-circuit rule.
Iteration cap: default max_iterations: 5. Configurable per-thread by writing { "max_iterations": <N> } to <thread>/.anvil.json. Exceeding the cap marks the thread BLOCKED and requires human review.
Command dispatch
| Command | Role | Reads | Writes |
|---|
ip-uspto | portfolio orchestrator | all <thread>.* dirs under cwd | (none; reports state per thread + recommends next command) |
ip-uspto-intake <thread> | intake | inventor disclosure (transcript, brain dump, notes) in <thread>/refs/ | <thread>/BRIEF.md (structured) |
ip-uspto-inventorship <thread> [--evidence [<repo>]] [--reseed] | inventorship interviewer | <thread>/BRIEF.md, latest <thread>.{N}/claims.tex if present; with --evidence also the implementation repo's git history (via the promoted anvil/lib/inventorship_evidence.py) | <thread>/inventorship.md (matrix); with --evidence also <thread>/inventorship-evidence/ (map + evidence.jsonl; Notes-column citations only) |
ip-uspto-pre-flight <thread> | pre-flight checker | latest <thread>.{N}/ (all files) | <thread>.{N}.preflight/ with _summary.md, findings.md, _meta.json |
ip-uspto-draft <thread> | drafter | <thread>/BRIEF.md, <thread>/inventorship.md, <thread>/refs/, <thread>/prior-art/; for revisions also prior version + all critic siblings | <thread>.{N}/ with spec/claims/abstract/drawings |
ip-uspto-review <thread> | general reviewer | latest <thread>.{N}/ | <thread>.{N}.review/ |
ip-uspto-101 <thread> | §101 critic | latest <thread>.{N}/ | <thread>.{N}.s101/ |
ip-uspto-112 <thread> | §112 critic | latest <thread>.{N}/ | <thread>.{N}.s112/ |
ip-uspto-claims <thread> | claims critic | latest <thread>.{N}/claims.tex + <thread>.{N}/spec.tex | <thread>.{N}.claims/ |
ip-uspto-prior-art <thread> | prior-art critic | latest <thread>.{N}/ + <thread>/prior-art/** | <thread>.{N}.priorart/ |
ip-uspto-adversary <thread> | adversarial critic (optional, opt-in via .anvil.json) | latest <thread>.{N}/ + <thread>/prior-art/** (optional) | <thread>.{N}.adversary/ (findings-only — all nine dims null) |
ip-uspto-fto <thread> | FTO triage critic (optional, on-demand; .anvil.json opt-in also supported) | latest <thread>.{N}/ + <thread>/fto-refs/** (required) | <thread>.{N}.fto/ (report-only — all nine dims null, never flags; NOT an FTO opinion) |
ip-uspto-vision <thread> | drawing vision critic (optional) | rendered drawings under latest <thread>.{N}/drawings/ (SVG/PNG; drawings only — never the spec PDF) | <thread>.{N}.vision/ with _review.json (kind=vision) |
ip-uspto-revise <thread> | reviser | latest <thread>.{N}/ + ALL <thread>.{N}.<tag>/ critic siblings | <thread>.{N+1}/ with _revision-log.md |
ip-uspto-audit <thread> | auditor | READY <thread>.{N}/ | <thread>.{N}.audit/ |
ip-uspto-figures <thread> | figurer | latest <thread>.{N}/spec.tex + reference numerals | <thread>.{N}/drawings/** |
ip-uspto-finalize <thread> | finalizer | AUDITED <thread>.{N}/ + <thread>/inventorship.md | <thread>.final/ with submission package |
The portfolio orchestrator is the user-facing entry point for status; the lifecycle commands are dispatched from it (or invoked directly by the orchestrating agent).
Multi-critic primitive — sibling directory convention
Given an artifact at <thread>.{N}/, critic outputs land in sibling directories with the same parent and name prefix:
<thread>.{N}/ ← the artifact (immutable once review starts)
<thread>.{N}.review/ ← general reviewer
<thread>.{N}.s101/ ← §101 critic
<thread>.{N}.s112/ ← §112 critic
<thread>.{N}.claims/ ← claims critic
<thread>.{N}.priorart/ ← prior-art critic
<thread>.{N}.adversary/ ← adversarial critic (optional, opt-in; findings-only — all nine dims null, critical-flag eligible)
<thread>.{N}.fto/ ← FTO triage critic (optional, on-demand; report-only — all nine dims null, NEVER flags; triage-for-counsel, not an FTO opinion)
<thread>.{N}.vision/ ← drawing vision critic (optional; kind=vision, scores rendered drawings only)
<thread>.{N}.preflight/ ← pre-flight (mechanical compliance) — produced after revise, pre-review
<thread>.{N}.audit/ ← final fact-check (audit phase, post-convergence only)
<thread>.{N+1}/ ← reviser output (consumes ALL siblings above)
Naming rule: <thread>.{N}.<tag>/. The <tag> is a single short token; no nesting, no dots within the tag. Discovery is "glob <thread>.{N}.*/ minus the bare <thread>.{N}/".
Uniform critic output schema
Every critic directory contains:
<thread>.{N}.<tag>/
_summary.md Scorecard (9-dim /45 partial — critic only fills dimensions it owns) + critical flag boolean
findings.md Itemized findings, each with: severity, location (file:section), rationale, suggested fix
_meta.json { critic: <tag>, role: <which role md>, started: <iso>, finished: <iso>, model: <hint>, schema_version: 1, scorecard_kind: "machine-summary" }
Uniform schema enables ip-uspto-revise to enumerate findings programmatically without per-critic special-casing. Critics that don't fill a rubric dimension leave it null rather than zero — the reviser aggregates non-null scores by mean.
Schema note: this schema (_summary.md / findings.md / _meta.json) is the canonical machine-summary scorecard kind documented in anvil/lib/snippets/scorecard_kind.md. The memo, pub, slides, and report skills use the human-verdict kind (verdict.md / scoring.md / comments.md); the deck skill is the layered/aggregator reference (both kinds present). The two-kind discriminator (set in _meta.json as scorecard_kind) is how consumers distinguish the shapes without hardcoding skill-specific knowledge — see anvil/lib/snippets/scorecard_kind.md and anvil/lib/snippets/critics.md for the aggregation rules.
Reviser composition
ip-uspto-revise discovers critic siblings, aggregates their scorecards, and either advances or produces the next version. See commands/ip-uspto-revise.md for the full algorithm.
Key design property: critics are independent and parallelizable. The reviser is the synchronization point. Adding a new critic = adding a new ip-uspto-<critic>.md command + a new sibling tag. No reviser code changes.
Convergence loop
Lifecycle for one revision pass:
DRAFTED → (run all critics) → REVIEWED → (revise consumes ALL siblings) → REVISED → (pre-flight) → loop until convergence → READY → AUDITED → FINALIZED
The default critic set is review + s101 + s112 + claims + priorart. Operator can subset by writing { "critics": ["review", "s101", "s112", "claims"] } to <thread>/.anvil.json (e.g., skip priorart if no prior art was supplied; the reviser refuses to advance without all configured critics present).
Optional adversarial critic (issue #434): the adversary critic (commands/ip-uspto-adversary.md) is opt-in, not default — operators enable it by adding "adversary" to the critics array in <thread>/.anvil.json. It attacks the application (§103 obviousness combinations over supplied prior art + AAPA, design-arounds, §112(a) enablement-hole challenges) rather than verifying it, and is findings-only: all nine rubric dimensions stay null, so the aggregator's mean-of-non-null rule is unaffected; its critical flags short-circuit the verdict like any other critic's. Once configured, the reviser's all-configured-critics-present rule applies to it as-is.
Optional FTO triage critic (issue #446): the fto critic (commands/ip-uspto-fto.md) is on-demand, not default — typically run pre-finalize or before a non-provisional conversion; .anvil.json opt-in (add "fto" to the critics array) is also supported with the adversary's exact mechanism. It screens operator-supplied third-party references in <thread>/fto-refs/ (never performing its own patent search — a dedicated input dir, distinct from prior-art/, because FTO targets may postdate priority) on a 0–4 relevance scale, producing a structured triage-for-counsel report — it is NOT an FTO opinion and says so verbatim in both output artifacts. Like the adversary it is findings-only (all nine dims null), but with one departure: it is report-only and NEVER flags — critical_flag is always false, so an fto sidecar never blocks convergence; severity routes through counsel-action buckets instead. The reviser MAY consume its design-around vectors as claim-ladder additions; it never consumes a verdict from it.
Critic concurrency in v0: critics may be run serially or in parallel. The orchestrator (ip-uspto.md) reports "all configured critics done at version N" as a boolean — it does not enforce concurrency. Parallel spawn is a future enhancement that will land in anvil/lib/critics.py (issue #10); v0 implementations should default to serial for debuggability.
Progress tracking
Each <thread>.{N}/ directory contains _progress.json recording phase state. Schema:
{
"version": 1,
"thread": "<thread>",
"phases": {
"draft": { "state": "done", "started": "2026-05-28T14:00:00Z", "completed": "2026-05-28T14:30:00Z" },
"figures": { "state": "in_progress", "started": "2026-05-28T14:35:00Z" }
},
"metadata": {
"iteration": 1,
"max_iterations": 5
}
}
Phase states: pending, in_progress, done, failed. Validation is by file existence (does spec.tex exist? does _summary.md parse?), not by flag — _progress.json is a resume hint, not the source of truth. A phase that crashed mid-write should be re-runnable from pending after deleting any partial output.
The canonical _progress.json schema, read-merge-write recipe, and crash recovery contract live in anvil/lib/snippets/progress.md (in an installed consumer repo: .anvil/lib/snippets/progress.md); every command in this skill follows that convention. The merge is shallow: command updates one phase, preserves all others. All ip-uspto critic siblings (<thread>.{N}.review/, .s101/, .s112/, .claims/, .priorart/, .audit/, .preflight/) follow the machine-summary scorecard kind per anvil/lib/snippets/scorecard_kind.md: each emits _summary.md + findings.md + _meta.json (with scorecard_kind: machine-summary); each fills only its owned rubric dimensions and leaves others null for the reviser's mean aggregation.
Outline control surface
Each <thread>.{N}/ directory contains _outline.json — a typed control surface that records the section-by-section drafting plan for this version. The outline is load-bearing: it is the diff-able interface where an operator can inspect and edit the structure of the application before the drafter pays for full section generation, and it is the per-section resume index during drafting and revising.
The outline is additive to the existing draft outputs (spec.tex, claims.tex, abstract.txt, drawings/drawing-descriptions.md) — it does not replace them. Each outline section carries the file routing and heading macro the drafter uses to deterministically place its rendered output without re-deriving from the section id.
Schema
{
"schema_version": 1,
"thread": "<slug>",
"title": "...",
"iteration": 1,
"sections": [
{
"id": "field",
"file": "spec.tex",
"heading_macro": "\\fieldoftheinvention",
"target_tokens": 120,
"key_points": ["..."],
"sources_to_cite": [],
"status": "pending"
},
{
"id": "background",
"file": "spec.tex",
"heading_macro": "\\background",
"target_tokens": 1200,
"subsections": [
{"id": "problem", "key_points": ["..."]},
{"id": "prior-approaches", "key_points": ["..."]}
],
"sources_to_cite": ["doi:...", "arxiv:..."],
"status": "pending"
},
{
"id": "summary",
"file": "spec.tex",
"heading_macro": "\\summary",
"target_tokens": 800,
"key_points": ["mirror independent claim 1 at higher level", "..."],
"status": "pending"
},
{
"id": "brief-description-of-drawings",
"file": "spec.tex",
"heading_macro": "\\briefdescriptionofdrawings",
"target_tokens": 200,
"figures": [
{"n": 1, "caption": "..."},
{"n": 2, "caption": "..."}
],
"status": "pending"
},
{
"id": "detailed-description",
"file": "spec.tex",
"heading_macro": "\\detaileddescription",
"target_tokens": 6000,
"subsections": [
{
"id": "feature-1",
"feature_ref": "BRIEF.md#3.1",
"key_points": ["..."],
"ranges": [{"param": "freq", "range": "5GHz-80GHz", "preferred": "40GHz"}],
"alternatives": [{"param": "substrate", "values": ["Si", "GaAs", "InP"]}],
"refnums": [10, 12, 14, 16],
"target_tokens": 1800
}
],
"status": "pending"
},
{
"id": "claims",
"file": "claims.tex",
"target_tokens": 3000,
"claim_tree": [
{"n": 1, "type": "independent", "topic": "apparatus", "key_limitations": ["..."]},
{"n": 2, "type": "dependent", "parent": 1, "topic": "...", "drawn_from": "feature-1#alt:Si"},
{"n": 9, "type": "independent", "topic": "method", "key_limitations": ["..."]}
],
"status": "pending"
},
{
"id": "abstract",
"file": "abstract.txt",
"target_tokens": 200,
"word_cap": 150,
"status": "pending"
}
]
}
Field semantics
schema_version: integer, currently 1. Migrations bump this.thread: the thread slug; matches _progress.json.thread.title: human title (from BRIEF.md frontmatter).iteration: integer matching _progress.json.metadata.iteration. Bumped when the reviser copies the outline forward.sections: ordered array; the drafter MUST iterate in array order. The order is the authoritative render order. The minimum required section ids are field, background, summary, brief-description-of-drawings, detailed-description, claims, abstract.- For each section:
id: unique within the array. Maps onto the §5a–§5i drafter steps.file: target file (spec.tex | claims.tex | abstract.txt). Lets the drafter route rendered output deterministically without per-id special-casing.heading_macro: LaTeX macro that opens the section in spec.tex (omitted for claims.tex / abstract.txt, which use their own structure).target_tokens: drafter budget hint. Soft cap; the drafter MAY exceed if the inventive material justifies, but should report the overrun in its closing summary.key_points / subsections / figures / claim_tree: section-specific structured content the drafter conditions on. Free-form within their typed shape.sources_to_cite: optional citation identifiers (DOI, arXiv, USPTO publication number). Slot for future citation primitive.status: lifecycle state, see below.
Status lifecycle
Per-section status values mirror _progress.json phase states:
| State | Meaning |
|---|
pending | Section has not been rendered yet for this version. |
in_progress | Section render started but did not complete (crash, abort). |
done | Section has been rendered; its bytes in file are valid. |
failed | Section render attempted and failed (error captured in _progress.json.phases.draft.errors). |
The drafter advances a section from pending → in_progress → done (or failed) one section at a time, persisting _outline.json after each transition so a crash leaves a recoverable state.
Validation rule
Consistent with the rest of the skill: file existence and section presence in the target file win over the flag. A section flagged done whose bytes are absent from file is treated as not-done (re-rendered on resume). A section flagged pending whose bytes ARE present in file (because, say, the operator hand-wrote it) is treated as done (skipped). The flag is a resume hint; the file is the source of truth — same rule as _progress.json.
The draft phase is done only when every section has status: done AND its bytes validate by the file-existence check.
Schema location
The schema is documented inline here for v0. There is no separate schemas/ directory; promotion to a versioned JSON Schema file is deferred to anvil/lib/ extraction under issue #10.
Rubric
See rubric.md for the 9-dimension /45 USPTO scoring schema, the ≥39 advance threshold, and the §101/§112 critical-flag short-circuit policy. The optional ip-uspto-vision critic owns a separate drawing-vision rubric subset (dv1–dv5, /25) documented in the same file — it critiques the rendered drawings only (legibility, line weight/contrast, label placement, figure-number visibility, cross-reference accuracy) and ships its scorecard directly as _review.json (canonical kind=vision schema) rather than the _summary.md/findings.md machine-summary shape the source-side critics use; both are discovered and aggregated uniformly by anvil/lib/critics.py.
Project BRIEF artifact type
ip-uspto is registered as a skill-identity artifact_type value in
the shared project-BRIEF registry
(anvil/lib/project_brief.py::REGISTERED_ARTIFACT_TYPES /
SKILL_IDENTITY_ARTIFACT_TYPES; issue #440, following the
#386/#408/#432 pattern for deck/slides/proposal/pub/report).
In a shared project BRIEF, a documents: entry with
artifact_type: ip-uspto declares that this skill owns the thread. It
is NOT a memo subtype: it selects no memo rubric overlay, and memo
commands fail loudly when pointed at an ip-uspto-declared thread.
anvil:project-migrate's letter-family adoption mode
(--adopt-family) writes this value (with a # TODO(operator)
confirmation marker) when the operator passes the REQUIRED
--artifact-type ip-uspto — there is no inference between a full
application and a provisional (ip-uspto-provisional), so the choice
is always explicit.
USPTO-specific phases
Beyond the standard draft → review → revise → figures → audit lifecycle, this skill adds four USPTO phases:
| Phase | Command | When | Purpose |
|---|
| Intake | ip-uspto-intake | Before first draft | Convert raw inventor disclosure into a structured brief: problem, prior approaches, key inventive features, embodiments, ranges, edge cases. Without this, the drafter hallucinates. |
| Inventorship | ip-uspto-inventorship | Before first draft; re-checked pre-finalize | Generate inventor interview prompts to attribute each independent claim concept to ≥1 named inventor. 37 CFR 1.63 inventor oath requires correct inventorship; mis-attributed inventorship is grounds for unenforceability. Opt-in --evidence mode mines the implementation repo's git history (the promoted shared anvil/lib/inventorship_evidence.py, pure stdlib + subprocess git) into reduction-to-practice citations that pre-fill the matrix Notes column only — advisory evidence for the attorney interview; conception attribution and the ● rules are untouched. |
| Pre-flight | ip-uspto-pre-flight | After each revise, before next review | Mechanical compliance scan: paragraph numbering ([0001], [0002], ...), abstract word count ≤150, claims numbered 1..N, no multiple-dependent-on-multiple-dependent claims (37 CFR 1.75(c)), margin/font checks via LaTeX class, render-gate (compile + overfull-box + source-side placeholder scan via anvil/lib/render_gate.py — the LaTeX-skill analog of marp_lint; page_cap=None since patents are uncapped; consumers can override per-thread via <thread>/.anvil.json: render_gate.page_cap; call-site overfull_threshold_pt=2.0 override per the legal-artifact calibration, issue #572 — tighter than the framework default 5.0pt). Deterministic-first with LLM fallback for ambiguous cases. Render-gate is mechanical pass/fail (Check 9, no rubric score) — failure short-circuits pre-flight per the standard rule. See commands/ip-uspto-pre-flight.md Check 9. |
| Audit | ip-uspto-audit | After convergence, before finalize | Adds a render-gate backstop (issue #572, Check 11): re-invokes compile_and_gate(...) at overfull_threshold_pt=2.0 so a late-revise overfull box introduced AFTER the last pre-flight pass cannot reach FILING-READY unchallenged. The result is written to the audit sibling's _gate.json, which ip-uspto-finalize's pre-finalize gate reads. |
| Finalize | ip-uspto-finalize | After AUDITED | Assemble submission package: spec.pdf, drawings.pdf, ADS placeholder, fee schedule placeholder, inventorship attestation. Pre-finalize gate (issue #572): reads the audit sibling's _gate.json and refuses to assemble <thread>.final/ when any overfull-box finding is present. Does not file — that is a human + Patent Center action. |
Render-gate threshold calibration
The ip-skill pre-flight + audit-backstop call sites pass overfull_threshold_pt=2.0 to compile_and_gate(...), tighter than the framework default of 5.0pt in anvil/lib/render_gate.py. Rationale (issue #572): a filed provisional shipped with a 83.6pt overfull (~16× the framework default; >40× the ip-skill override). The 2.0pt call-site value is the legal-artifact calibration — strict enough to catch margin-breaking content well below the issue body's "egregious / >10pt" line, loose enough that sub-point cosmetic slop still passes. The framework default in render_gate.py remains 5.0pt to avoid disturbing the installation, proposal, datasheet, pub, report consumers that inherit it. The override is per-call-site, not per-skill-config, so an audit done outside the ip-skill commands (e.g., a consumer custom critic) does NOT inherit the tighter threshold unless it passes the same kwarg.
Defaults and overrides
This skill ships with opinionated defaults. Consumers extend liberally via .anvil/skills/ip-uspto/ in their own repo:
voice.md (optional) — Firm or attorney voice/style guidance the drafter reads in addition to its base prompt.rubric.overrides.md (optional) — Add domain-specific critical-flag examples; cannot reduce the base rubric.BRIEF.md.example — Reference brief shape; the intake command produces this shape from a disclosure.critics/ (optional) — Add custom critic command files (e.g., ip-uspto-mydomain.md). The orchestrator picks them up automatically by glob.
Important caveats
- This skill does NOT file a patent application. It produces a submission-ready package. Filing requires human review, attorney sign-off, and submission via USPTO Patent Center.
- This skill does NOT replace a licensed patent attorney. It is a drafting and review aid. Inventorship attestation (37 CFR 1.63), assignment, and prosecution strategy require a qualified human attorney.
- The prior-art critic does NOT do its own patent search. Operator must supply prior art in
<thread>/prior-art/. Patent search is a separate role potentially shipped as a future skill.
- The FTO triage critic produces triage-for-counsel, NOT an FTO opinion.
ip-uspto-fto screens only operator-supplied references in <thread>/fto-refs/ (it never searches), renders no infringement or clearance conclusion (0–4 relevance scores + counsel-action buckets are its entire vocabulary), and never marks output as privileged. Licensed patent counsel must validate every finding before any business reliance.
- Provisional applications are handled by the sibling skill
anvil:ip-uspto-provisional (issue #433): claims-optional posture, enablement-depth-dominant anvil-ip-provisional-v1 rubric (/45, ≥39; dim 9 Conversion readiness replaces this skill's Claim-spec correspondence), shared anvil-uspto.cls + intake substrate. The natural flow is provisional thread → (≤12 months) → a non-provisional conversion thread in THIS skill referencing it. Design patents remain out of scope — track as separate issues.
Conversion linkage — referencing a provisional (issue #501)
A non-provisional thread in this skill converts an earlier anvil:ip-uspto-provisional filing through a mechanical §119(e) linkage (no manual priority-claim drafting required):
- Declaration: the non-provisional
<thread>/BRIEF.md carries an optional converts_provisional frontmatter block (thread / filing_date / application_number / optional portfolio_path) — see commands/ip-uspto-intake.md §"converts_provisional". This is structured filing data, NOT a refs/ body-citation; cross_thread_refs.py is portfolio-root-relative and does not resolve the cross-portfolio case, so the BRIEF key (with portfolio_path for cross-portfolio) is the declaration surface.
- §119(e) priority-claim text is emitted from that block at two points: a "CROSS-REFERENCE TO RELATED APPLICATIONS" paragraph into
spec.tex at draft (ip-uspto-draft.md §5a), and the ADS domestic-priority data at finalize (ip-uspto-finalize.md step 10).
- 12-month deadline: the orchestrator (
ip-uspto.md step 5) computes filing_date + 12 calendar months via lib/conversion_deadline.py and warns loudly when within 60 days of (or past) the §119(e) deadline.
- Fail loud: a
converts_provisional block present with a missing/empty filing_date is an error at draft, finalize, and orchestrator surfacing — never a silently blank priority claim. The authoritative producer copy of the date is the provisional thread's _filing.json (written by ip-uspto-provisional-finalize); the BRIEF key is the consumer copy.
- §112(a) disclosure-coverage check (issue #517): beyond the mechanical date+boilerplate linkage above, the
s112 critic (commands/ip-uspto-112.md) gains a converts_provisional-gated §112(a) conversion disclosure-coverage check. When the block is present, the s112 per-claim-limitation support sweep is re-run with the provisional spec.tex as the baseline (resolved via converts_provisional.thread + optional portfolio_path at highest-N): does the provisional disclose each converted claim limitation at §112(a) written-description-and-enablement depth? Findings are advisory FOR COUNSEL — they flag possible new-matter / unsupported converted subject matter (claim-number + provisional-spec-paragraph citations) and never adjudicate priority (never declare priority lost, claim invalid, or conversion failed). An unsupported converted independent-claim limitation is critical-flag eligible; dependent-only / narrowly-supported gaps are non-critical. The check is fail-loud: if the provisional spec.tex cannot be resolved it emits a flagged "could not be performed" finding, never a silent pass. It is dormant / byte-identical when converts_provisional is absent — no new critic, no new orchestrator critic-set entry, no Python lib module, no new rubric dimension (total stays /45); the check rides on existing dim 2 §112(a).
Git sync hook (opt-in, off by default)
Consumers running anvil under an external orchestrator (a sphere channel-agent, a Loom-style daemon) can opt in to a per-phase git commit hook so every lifecycle phase leaves the working tree clean: a repo-level .anvil/config.json with git.commit_per_phase: true (and optionally git.push: true) has each write-bearing ip-uspto command end its phase by staging only the dirs it wrote and committing as anvil(ip-uspto/<phase>): <thread>.{N} [<state>]. The full contract — knob shape, defaults-off rule, commit-message format, staging scope, warn-and-continue failure semantics, and ordering after the _progress.json done write and the #350 sidecar atomic rename — lives in anvil/lib/snippets/git_sync.md (.anvil/lib/snippets/git_sync.md in an installed consumer repo). All 16 write-bearing ip-uspto commands adopt it; the read-only ip-uspto portfolio orchestrator is exempt by definition. When .anvil/config.json is absent or the knob is false, behavior is byte-identical to a pre-#426 install — the hook is default off.
