bug, error, broken |
bug |
screenshot/image attached + this screen is broken, fix what's in this screenshot, see attached screenshot, looks wrong (web) |
bug — image branch (routing-matrix § BUG Phase Contract → IMAGE BRANCH): five-section analysis at CLASSIFY, Visual Fix Loop at VERIFY. Native mobile screen vs reference image → MOBILE_NATIVE visualloop row. Screenshot + improve/make this better is overloaded → REDIRECT: visual/look-and-feel → restyle; feature behavior → kaizen; usability findings only (no code) → echo direct |
feature, implement, build |
feature |
security, vulnerability, CVE |
security (mode=to-zero on fix every finding the scanner reports, clear the security backlog, drive vulnerabilities to zero, get the SAST/SCA report to clean — a frozen-scope sweep at a CVSS floor per _common/FINDING_LEDGER.md; RISK-ACCEPTED requires a named owner + expiry, CRITICAL/HIGH acceptance is Ask First, lapsed acceptances re-open. One named vuln → the single-pass phases) |
refactor, clean up, code smell |
refactor (single known restructure, shipped once. Add an external review loop — "refactor until the reviewer is clean" — → quell profile=refactor, which keeps refactor's behavior-preservation invariant as an Equivalence Gate) |
optimize, slow, performance, speed up, make it faster, too slow, latency, p95/p99, slow query, slow endpoint, page load is slow, reduce memory, high CPU, throughput, bottleneck, slow build, slow CI, bundle size, cold start, over-fetching, chatty API |
optimize (any slow layer — code/render · query · network/payload · build/bundle · infra/runtime · search; layer located before a tool is bound. overloaded → REDIRECT: memory leak/OOM over time/unbounded growth → bug — a leak is a defect, not a tuning target; a steady-state footprint reduction → optimize. it got slower after deploy/regressed after release → optimize +Trail (bisect) when correctness holds, bug when it doesn't. Continuous parameter self-tuning (GC/threadpool/pool/cache-size loop) → AUTO_TUNING. mode=to-zero on fix every budget violation, get every route under its performance budget, clear the performance budget violations, all endpoints under their SLO, every bundle under its size budget — a set of budget violations swept to zero rather than one target to one number; budgets are frozen at BASELINE, so BUDGET-RAISED is a blocking finding only the adjudicator may ratify. One slow endpoint with one number → the single-pass phases. Contract → reference/optimize-recipe.md) |
kaizen, improve, polish, enhance existing, incremental improvement, refine, continuous improvement |
kaizen (UI-scoped → REDIRECT: polish/improve/enhance a feature → kaizen; polish the UI/improve the look and feel (UI/visual/look-and-feel scoped) → restyle row. Contract → reference/inline-recipes.md) |
anneal, design audit, audit the design, brush up the design, brush up the codebase, harden the design, harden the architecture, tighten the design, tighten loose ends, firm up the design, design weaknesses, design debt sweep, clean up the architecture, design improvement, thoroughly brush up the design, fix under-specified design, spec-code drift cleanup, align spec and code |
anneal (design improvement/improve the design alone is overloaded on the code-vs-visual axis → REDIRECT: architecture/code design → anneal; UI/visual/look-and-feel → restyle. Discovery-first, behavior-preserving design-quality audit → prioritized brush-up. spec-code drift splits on register-vs-fix: a report of every divergence, no edits → verity; align the code and clean it up → anneal (verity→anneal is the audit→fix pair). Disambiguation: apply a known restructure to a known target → refactor; a perf number → optimize; improve one feature vs a quantified target → kaizen; loop control → converge; no code, evolution directions → delve; remove features → trim; PR-gated quality-max → summit/acceptance; author a new spec → spec; report-only conformance → attest/pdm. Contract → reference/anneal-recipe.md) |
restyle, redesign, redesign the screen, UI refresh, visual refresh, visual polish, modernize the UI, make it look better, improve the look and feel, design facelift, polish the UI, improve the UI design, update the visual design, restyle the page |
restyle (UI/visual/interaction design improvement of an existing surface; no target → whole-UI sweep (confirm-before-launch, top-N by friction × visibility). Disambiguation: code/architecture design → anneal; one feature vs a metric target → kaizen; new capability → feature; direction only → vision direct; one named fix → palette/flow/muse direct; full design-to-impl pipeline → atelier skill; LP → funnel premium; drive an already-directed surface to zero findings from an external review engine → burnish (restyle → burnish is the pair). Contract → reference/restyle-recipe.md) |
review, check, audit (bare, overloaded — REDIRECT by object) |
record coherence (docs vs code, stale info, unexplained artifacts) → verity; a change set vs the standing decision record (ADR/spec/contract/convention) → abide — the split from verity is the anchor, not the subject: no change anchor → verity, "audit the recent changes / this PR / this release against the docs" → abide; code/PR quality → judge direct; skill/plugin/MCP supply chain → SUPPLY_CHAIN_AUDIT (Chain); active malware/infection → Chain; design fidelity vs mockup → DESIGN_AUDIT (Pixel); ecosystem health → ECOSYSTEM (Darwin→Gauge); standards (OWASP/WCAG/OpenAPI) → Canon; SKILL.md normalization → Gauge; a supplied prompt's residual vagueness → PROMPT_SPEC (Chisel audit) |
design system docs, token docs, component catalog |
DESIGN_SYSTEM_DOCS (see Routing Quick Start) |
brainstorm, bounce ideas, riff, ideate, sounding board |
(Flux direct — single-agent) |
apex, auto-impl, full implementation, discovery to launch, end-to-end feature, ultimate |
apex |
deliver, build a product from scratch, build an MVP, MVP delivery, product lifecycle, ship the product, scope-adaptive delivery, build-first, time to first PR |
deliver (scope-adaptive product/MVP delivery through the minimum viable specialist chain. Disambiguation: one bounded capability → feature; high-investment fixed deep discovery→ship run → apex; execute an existing Charter → enact. Contract → reference/deliver-recipe.md) |
charter, instruction document, team charter, team operating manual, runbook for a team, analyze repo and design a team, self-driving team charter, team design spec, comprehensive repo analysis to a plan |
charter |
enact, run the charter, execute the charter, execute the instruction document, build team from charter and run, orchestrate the charter, run docs/CHARTER.md |
enact |
layer, operating layer, project operating layer, project skill suite, set up the project's agents/recipes/workflows, design this repo's tooling, repo task playbooks, project routing map, design the project's agents and recipes together, give this repo a skill suite |
layer (Sigil[blueprint] designs the operating-layer blueprint, Sigil authors bodies, and Nexus registers routing; design-only stops at the blueprint. Disambiguation: one delivery's team+work plan → charter; one skill body → sigil/SKILL_GEN; one feature spec → spec. Contract → reference/layer-recipe.md) |
goal, /goal setup, goal recipe, long-running goal, autonomous loop setup |
goal |
gedanken, thought experiment, what if, reason through, counterfactual, limiting case, reductio ad absurdum, intuition pump, steelman scenario, hypothetically |
gedanken (structured thought-experiment reasoning — hypothetical → independent lenses → controlled perturbation → adversarial refutation → a falsifier; no code. Intent-anchored, not keyword-anchored: verdict-seeking (wants a final go/no-go) → magi direct; process-oriented (wants the reasoning walked through, no decision demanded) → gedanken. Worked example: "think through whether microservices are worth it here, no code" → gedanken (process-oriented framing overrides the "worth it" phrasing that would otherwise suggest magi). Exploratory analog of magi; orchestrates flux/magi/omen/matrix. Trivial one-off "what if" → flux/magi direct. Contract → reference/gedanken-recipe.md) |
delve, deep dive, deep-dive a feature, dig into the feature, dig deeper into <feature>, unlock the feature's potential, evolve this feature, deepen this feature, feature evolution, what more can we do with <feature> |
delve (existing-feature deep-dive → evolution-direction dialogue — excavate a shipped feature to non-obvious insights → evolution directions (deepen/broaden/reframe); no code, stops at a named Evolution Map, hands off to spec/kaizen/feature/apex. Disambiguation: executes improvement vs a target → kaizen; shapes a new idea → spec; abstract hypothesis → gedanken; which-feature verdict → essential/killer/trim. Trivial "what could we do with X" → flux/spark direct. Contract → reference/delve-recipe.md) |
spec, spec out, spec this out, flesh out the spec, nail down the requirements, write the spec with me, talk through a feature, refine until the spec is locked, idea to spec, finalize the spec, requirements through dialogue |
spec (interactive feature-proposal → locked spec via deep dialogue; INTERACTIVE default, no code; hands off to feature/apex. Disambiguation: which-feature verdict with minimal dialogue → essential/killer; builds → feature/apex. Contract → reference/spec-recipe.md) |
cartograph, reverse-engineer across repos, reverse the codebase, bird's-eye view, bird's-eye diagram, overview diagram, fukanzu, architecture map, architecture overview, map the system, design doc from code, document how the system works, understand the system across repos, cross-repo architecture, how do these repos fit together, system-wide diagram |
cartograph (multi-repo reverse-engineering → bird's-eye architecture diagrams + design doc; no code. Params: repos=a,b,c. Disambiguation: one shipped feature → evolution dialogue → delve; one repo → team+work plan → charter; plan-vs-code status → pdm; black-box external → rebuild → clone; changes the system → migrate; domain-preset doc generation → package; single repo / one diagram → lens/atlas/canvas direct. Temporal sibling → chronicle. Contract → reference/cartograph-recipe.md) |
chronicle, repository history, repo history, commit history summary, summarize the commit history, history of the repository, history of the repo, how did we get here, evolution of the codebase, codebase evolution, project timeline, repo timeline, git history narrative, git history summary, narrative changelog, story of the repo, retrospective from git, feature history, bug/defect history, improvement history, decision history, decision log from history, how the decisions were made, ADR from git history, design philosophy from history, project ethos, project worldview, what does this project value |
chronicle (commit-history reverse-engineering → era timeline + storylines + decision log + per-lens deep-dives + inferred ethos; no code. Params: since=<tag|date> · <path> · repos=a,b,c · lenses=<set|none>. Disambiguation: one period's PR report → launch direct; one diff to teach → tome direct; which commit broke X → trail direct; a new forward ADR → atlas direct; a live judgment call → magi direct; one feature → delve; repo → team+work plan → charter; plan-vs-code status → pdm. Spatial sibling → cartograph. Contract → reference/chronicle-recipe.md) |
verity, consistency audit, coherence audit, audit the codebase and the docs, investigate the code and docs thoroughly, docs vs code, the docs contradict the code, docs and code are out of sync, contradictions across the repo, list every inconsistency, list the discrepancies, outdated documentation, stale docs, what information is out of date, find what no longer applies, dangling references in the docs, the README is wrong, why does this exist, nobody knows why this is here, no rationale anywhere, undocumented decision, unexplained constant, magic number with no reason, unknown provenance, orphaned code nobody can explain, knowledge-transfer audit, before the maintainer leaves, hand-over audit |
verity (codebase × documentation coherence audit → a triaged Verity Register of three classes — CONTRADICTION · STALE · UNEXPLAINED; two mandatory sweeps (claims→reality, reality→record), authority adjudication per contradiction (UNDECIDED is a first-class outcome — never auto-resolved toward the code), a mandatory Provenance Search Record behind every UNEXPLAINED entry, root-cause clustering, and a route per finding. Report-only — no code, no doc edits. Disambiguation: find design weaknesses and fix them → anneal (verity→anneal is an audit→fix pair; bare spec-code drift splits on register-vs-fix); what is built vs planned → pdm direct; conformance to one given spec → attest; external named standards → canon skill; design-system token denominator → lattice; unused files to delete → sweep direct; one archaeology or comprehension question → trail/lens direct; a cluster of unexplained decisions → verity → chronicle → atlas; an article's claims vs the world → newsroom. Spatial sibling → cartograph; temporal sibling → chronicle. Contract → reference/verity-recipe.md) |
abide, ADR drift, does this change follow the ADR, did we drift from the ADR, changes vs the ADRs, recent changes vs the design docs, audit the recent changes against the docs, check the diff against the decision record, did this PR break a decision, we changed it but never updated the ADR, undocumented architecture decision, no ADR was written for this, decision record drift, governance audit, architecture decision compliance, saikin no henkou to ADR no kairi, ADR to no kairi wo kansa |
abide (change-anchored governance audit → a triaged Divergence Docket of four classes — VIOLATES · SUPERSEDES-SILENTLY · OBSOLETE-COMPLIANCE · UNGOVERNED; the governance corpus is frozen independently of the change set, so records the change never touched still bind; two mandatory join directions (rules→hunks finds violations, hunks→rules finds the ADR that was owed), an intent fork where UNDECIDED is first-class and never defaults to "deliberate", security/privacy-control rules are non-supersedable, and a mandatory Governance Search Record behind every UNGOVERNED entry. Report-only — never edits a record; SUPERSEDES-SILENTLY routes to a new superseding ADR via atlas, never an edit of the original. Disambiguation: the standing record vs the code repo-wide, no change anchor → verity (verity since= cannot see this recipe's core case — code moved, the ADR did not, so the ADR falls outside its touched inventory); the diff's bugs/quality → judge direct; commit/PR packaging → guardian direct; conformance to one given spec → attest direct; built-vs-planned → pdm direct; external named standards → canon skill; write the ADR itself → atlas direct (abide→atlas is the audit→author pair); proof-carrying merge evidence → acceptance. Docket sprawling repo-wide → abide→verity; an UNGOVERNED cluster → abide→chronicle→atlas. Contract → reference/abide-recipe.md) |
essential, must-have, MVP definition, core feature, minimum viable, cut scope, bare minimum |
essential |
killer, killer feature, differentiator, WOW experience, decisive feature, competitive edge |
killer |
trim, remove feature, dead weight, dead-weight feature, cut unused feature, feature sunset, retire feature, delete what we don't need, prune features, kill unused feature |
trim (inverse of essential/killer — removal via 2-axis filter; core engine void. Contract → reference/inline-recipes.md) |
acceptance, proof-carrying PR, acceptance gate, machine-adjudicated merge, tier-s merge, payment merge, auth merge, auto-merge with evidence |
acceptance |
growth-acceptance (alias of acceptance layer=c), lifecycle gate, market proof, research proof, brand proof, insight ledger, incrementality gate, brand compiler, growth-brand contract, post-launch measurement |
growth-acceptance |
summit, tri-engine, all engines, claude+codex+agy, quality maximization, strategic decision, release-critical, design-critical launch |
summit |
podium, slide deck, keynote, conference talk, presentation, talk deck, speaker deck, onboarding kit (doc + deck), learning material with companion deck, doc + slide, unified content package, article + slides, retrospective (doc + exec deck), launch package (announcement + sales deck) |
podium |
newsroom, fact-check, fact check the article, verify sources, verify the sources, source-grounded article, every claim cited, all claims sourced, no speculation, no guesswork, citation audit, citation check, check for misinformation, check this is actually correct, check the article is correct, grounded writing, evidence-backed article, provenance for every statement |
newsroom (grounding-first article production/audit — Evidence Ledger → producer≠verifier citation-support audit → adversarial refutation (≥2 independent T1–T3 sources) + staleness check → remediation ≤2 cycles; confirm release-critical; ships a named Provenance Report, zero ungrounded facts for external-facing, unresolvable claims marked [UNVERIFIED] never silently dropped. Disambiguation: doc+slide package with polish maximization → podium; no grounding requirement → tome direct; impl-vs-spec conformance → attest; named-standards compliance → canon. Contract → reference/newsroom-recipe.md) |
wish, once-in-a-lifetime request, favor of a lifetime, isshou no onegai, begging you, your absolute best, your very best output, spare nothing, pull out all the stops, no second chance, one shot to get this right, masterpiece quality |
wish (scarcity-gated one-shot quality-ceiling wrapper, deliverable-agnostic; ACCEPT = all rubric dims = 3; budget-enveloped, always confirm. Disambiguation: strategic/release-critical code quality-max wanting engine triangulation → summit; iterate to a standard bar (revisions possible) → converge; bare quality maximization without one-shot/lifetime framing → summit. Casual "best quality please" on routine work → the natural recipe, wish's own Scarcity Gate recommends the downgrade. Contract → reference/wish-recipe.md) |
runway, design tournament, flagship screen design, best possible design, design quality maximization, compete design directions, product-defining screen, ultimate UI design |
runway (flagship UI design tournament — 3 orthogonal directions [brand-led / usability-led / trend-led] → persona-panel judging → winner + salvage → ceiling convergence, ACCEPT = all rubric dims = 3; named Runway Board; always confirm. Disambiguation: single-direction improvement to a standard bar → restyle (its ✓direction-gate escalates here on tied directions or a product-defining surface); acquisition/conversion LP → marquee; direction-only, no execution → vision direct; strategic code quality-max → summit. Contract → reference/runway-recipe.md) |
hallmark, brand identity, brand book, brand voice, visual identity, define the brand, brand guidelines, brand core, create a brand |
hallmark (brand identity package quality-max — brand-core contract-level dialogue (values/voice/prohibitions) → identity tournament (Vision ×3 → Muse tokens + Ink icons) → persona-resonance panel + differentiation refutation per _common/ADVERSARIAL_REFUTATION.md; named Hallmark Charter; confirm-before-launch. Disambiguation: personal/engineer branding → compete; propagate a settled brand across surfaces → rebrand (create→propagate pair); ship-time brand verification → growth-acceptance; brand equity research → BRAND_EQUITY (Compete). Contract → reference/hallmark-recipe.md) |
rebrand, brand refresh, apply new brand everywhere, new logo everywhere, brand migration, propagate the brand, brand consistency sweep, roll out the new brand |
rebrand (all-surface brand propagation with a proven-complete guarantee — INVENTORY freezes the touchpoint denominator → per-surface batches under brand rubric + a11y + no-regression → RESIDUE-GATE (old tokens/voice grep + visual scan, 2× zero) → old-brand DECOMMISSION; named Consistency Attestation; Ask First on big-bang / 10+ files. Disambiguation: no settled Brand Book yet → hallmark first; one surface → restyle; one token change → muse direct; technical change-completeness → migrate. Contract → reference/rebrand-recipe.md) |
marquee (alias of wish domain=lp), best possible landing page, flagship LP, one-shot LP, LP of a lifetime, ultimate landing page, premium landing page, landing page masterpiece |
marquee (wish-grade one-shot LP production — crystallization dialogue → 3-direction tournament [narrative-led / design-led / proof-led] → funnel premium-craft-axes production → gauntlet + ceiling convergence with machine oracles (Lighthouse ≥95 / CWV green / WCAG AA) + LP-specific refutation → One-Shot Gate; ACCEPT = all rubric dims = 3; no Scarcity Gate — intentional vs wish (LP demand is legitimately repeatable); named Marquee Dossier; always confirm. Bare landing page is overloaded → REDIRECT: routine LP build → funnel premium skill / funnel; wish-grade one-shot/flagship → marquee (this recipe); improving an existing LP → kaizen/restyle; in-product flagship screen → runway. Contract → reference/marquee-recipe.md) |
assay, prove the design, prove this is a good design, is this actually a good design, is this abstraction necessary, do we still need this abstraction, is the complexity essential, accidental complexity, prove the architecture rules hold, are the layering rules respected, architecture fitness functions, test the design claims, prove changes stay local, how expensive is change here, is our design debt real |
assay (experimental proof of code/architecture design claims; apply=true is opt-in and never inferred. Disambiguation: discover undiagnosed weaknesses and fix them → anneal (assay→anneal is a measure→fix pair; anneal alone when the ask is "find what's wrong"); a known restructure → refactor; merge/PR-scoped gate → acceptance/summit; blast radius without implementing → ripple skill; propose-only YAGNI → void skill (judged vs experimented is the boundary); coverage/flaky → radar skill; spec↔impl conformance → attest/pdm; UI token conformance → lattice. Contract → reference/assay-recipe.md) |
crucible, prove it's usable, prove the UI works, stress the UI, stress test the interface, hostile conditions, adversarial conditions, works on a bad network, works offline, empty state and overflow, edge-case content, long labels break the layout, screen reader can complete it, keyboard-only, usable one-handed, does it hold up in the real world, operability |
crucible (operability floor proof, binary per-cell oracle. Disambiguation: craft/visual ceiling of a flagship surface → runway; standard-bar visual improvement → restyle; fails under ideal conditions → bug; system load/chaos/resilience → siege skill; named-standard conformance only → canon skill; one identified a11y fix → palette direct; E2E suite authoring → voyager direct. Contract → reference/crucible-recipe.md) |
silhouette, looks generic, looks templated, looks like every other SaaS, looks AI-generated, indistinguishable from competitors, would anyone know it's ours, nothing distinctive about it, blends in with the category, logo-off test, visual differentiation, stop looking like everyone else, stop looking templated |
silhouette (distinction proof for a product surface. Disambiguation: no settled brand to derive from → hallmark first; craft quality maximization → runway; standard-bar improvement → restyle; positioning research with no design change → compete skill; novel mechanism rather than a look → eureka; operability → crucible. Contract → reference/silhouette-recipe.md) |
lattice, design system drift, design system is not being used, is the design system actually used, token conformance, hardcoded colors everywhere, magic numbers in the CSS, one-off components, duplicate tokens, inconsistent spacing across screens, looks like five teams built it, design system audit, design consistency sweep |
lattice (design-system coherence proof, steady state; precondition: a system of record exists, else muse/vitrine first. Disambiguation: brand identity change propagation → rebrand; design the tokens → muse direct; catalog components → vitrine direct; improve how it looks → restyle/runway; code/architecture design weaknesses → anneal; dead-code removal → sweep direct; technical change completeness → migrate; between platforms → chorus. Contract → reference/lattice-recipe.md) |
chorus, cross-platform consistency, cross-platform coherence, iOS and Android feel different, web and app are out of sync, the apps drifted apart, native but the same product, platform parity, same product on every platform, does it feel native on both |
chorus (cross-platform coherence proof; precondition ≥2 shipped/specced platforms. Disambiguation: the porting plan → port skill (port→chorus is a plan→prove pair); implement one platform → native skill / MOBILE_NATIVE; drift within one platform's system → lattice (compose: lattice ×N, then chorus across); conditions rather than platforms → crucible; one flagship screen's ceiling → runway. Contract → reference/chorus-recipe.md) |
/Nexus (no arguments) |
proactive mode (non-Recipe) |
