HCJC

Use when editing web/static/style.css in the JCStream project. Covers the light-theme token system (--bg/--fg/--accent), the 10-tier .tier-F1…MM ladder palette, .ladder-* cells, body.is-table table-mode rules, content-visibility, mobile breakpoint at 720px, and the print at-rule. Trigger phrases: "restyle the cards", "fix the tier colors", "make the hero bigger", "add a CSS rule", "edit the stylesheet", "tweak the lightbox", "fix the table view", "recolor the F3 chip", "tune the print layout".

Use when auditing JCStream for accessibility — WCAG AA color contrast on the light theme, ARIA correctness (aria-current/aria-pressed/aria-modal/role=…), keyboard navigation, focus ring, tab order, alt text, screen-reader-only content, reduced-motion. The new light theme retired the dark theme's hand-tuned contrast tokens, so contrast must be verified empirically rather than assumed from tokens. Trigger phrases: "accessibility audit", "a11y", "a11y review", "WCAG check", "ARIA", "ARIA check", "screen reader", "contrast issue", "color contrast", "focus ring", "keyboard nav", "alt text".

Use when adding or modifying Python helper functions in web/build.py, web/classify.py, or web/shape.py that compute values for Jinja templates in the JCStream project. Covers the env.globals registration pattern (in web/build.py), the Inmate/Snapshot models, the _DEGREE_RE / _CHAPTER_LABEL / _OFFENSE_CATEGORY maps (now in web/classify.py), and the snapshot-shape helpers (now in web/shape.py). Trigger phrases: "add a helper for X", "compute Y per inmate", "expose Z to templates", "register a Jinja global", "compute bond/tier/category for inmate", "categorize this ORC code".

Use when reviewing the JCStream presentation layer — Jinja templates in `web/templates/*.html`, the RSS feed `web/templates/feed.xml`, the XSLT stylesheet `web/static/feed.xsl`, and the no-required-JS enhancement script `web/static/main.js`. Covers XSS-safe escaping (every `{{ }}` and `|safe`), inline JSON-LD shape, RSS validity, semantic HTML structure, XSLT correctness, progressive-enhancement contract (lightbox / filter / tier-tooltip degrade without JS), and third-party-script hygiene (none allowed per FCRA). **Read-only** — produces a finding report and hands fixes off to `jcstream-template-author` / `jcstream-a11y-auditor` / `jcstream-legal-copy-author`. Trigger phrases: "review the templates", "review inmate.html", "review the homepage", "check the RSS feed", "review the JSON-LD", "schema.org review", "review main.js", "review the filter JS", "review the lightbox", "audit the feed.xsl", "template code review", "XSS check", "Jinja escape audit".

Use when conducting a full code review of a JCStream PR, branch, or the whole repo. Orchestrator that fans out to the four language / domain reviewers in parallel — `jcstream-python-reviewer` (55.6% of code), `jcstream-template-reviewer` (29.3%), `jcstream-css-reviewer` (15.1%), `jcstream-security-reviewer` (cross-cutting compliance) — collects their findings, dedupes by file+line, ranks by severity, and emits a single PR-comment-ready consolidated review. **Read-only** — produces a single merged report; the underlying reviewers each produce their own per-domain reports as the trace. Trigger phrases: "code review", "review the whole project", "PR review", "review my changes", "review this branch", "review the diff", "full audit", "review everything", "comprehensive code review", "review PR".

Use when reviewing `web/static/style.css` for code-quality issues distinct from rendered accessibility. Covers dead rules, cascade conflicts, duplicate / self-overriding declarations, 10-tier ladder consistency (`.tier-F1`–`.tier-MM`, `.ladder-F1`–`.ladder-MM`), `body.is-table` table-mode parity, mobile breakpoint hand-offs (720px is the primary, 1024px / 1080px / 540px are secondary), content-visibility correctness, print at-rule completeness, unused CSS custom properties, hex duplication vs token use, and reduced-motion scope. **Read-only** — produces a finding report and hands fixes off to `jcstream-stylesheet-author` for code edits or `jcstream-a11y-auditor` for rendered-contrast / ARIA-affecting findings. Trigger phrases: "review the stylesheet", "audit style.css", "check the cascade", "review the print rules", "review the responsive breakpoints", "CSS code review", "find dead CSS", "find duplicate selectors", "review the tier ladder palette".

Use when reviewing Python code in the JCStream project — `scraper/*.py`, `web/build.py`, `web/classify.py`, `web/shape.py`, `tests/*.py`. Covers type hints, error handling at boundaries only (per CLAUDE.md), regex anchoring (post-`_DEGREE_RE` regression class), security (subprocess, requests `verify=`, eval/exec), dependency footprint, concurrency / thread safety in the sweep loop, stdlib idioms, dead code, and the project's no-print logging convention. **Read-only** — produces a finding report, hands fixes off to the relevant author skill. Trigger phrases: "review the python code", "review the scraper", "review web/build.py", "review classify.py", "audit the test suite", "code review the python", "lint check python", "type check the build helpers", "regex review", "thread-safety review", "python security review".

Use when modifying or extending data pullers in scraper/ — HCSO inmate roster (sweep.py), Cincinnati Open Data feeds (cfs.py, cfs_pdi.py, shootings.py, open_data_feeds.py), the courtclerk URL helpers, PRA email loop. Covers rate-limit budget, sweep health guards (sweep_guards.py: 50% floor, 10% error ceiling, detail watchdog, photo prune), atomic write to data/current.json, and the 30-min cron in .github/workflows/sweep.yml. Trigger phrases: "add a new data feed", "add an Open Data feed", "Socrata pull", "fix the HCSO scraper", "fix the sweep cron", "tune the sweep guard", "tune detail watchdog", "raise/lower roster guard", "PRA email loop", "courtclerk link helper", "photo prune skipped", "rate-limit", "sweep wall-clock".

Governs how Claude interfaces with this user's AuDHD cognitive architecture across every conversation, every domain, every mode. This skill fires on ALL interactions. It controls behavioral output: session starts, file uploads, retrieval, overwhelm response, conclusion validation, working memory externalization, and register calibration. Use on every conversation. If unsure whether it applies, it applies. Trigger on: any greeting, file upload, reference to prior work, confusion, frustration, overwhelm, drafting request, complex multi-step task, or any request referencing "the way I like it" or "you know how I work" or "based on what you know about me." Also trigger when the user provides minimal instruction and expects inference, or arrives with a conclusion wanting validation rather than reconstruction.

Audits cited legal authorities, regulatory references, and factual assertions for currency, accuracy, and correct application. Use when the user needs to verify statutes have not been amended or repealed, regulatory citations reflect current rules, factual claims match the supplied record, or a work product's legal foundation is reliable before filing. Also use when the user says "check the citations," "verify this is current," "audit the authorities," "is this statute still good," or asks about any specific ORC section, U.S.C. provision, or C.F.R. part. If the user is about to file, publish, or send anything with legal citations, this skill applies.

Verifies claims, quotes, timelines, and findings against supplied source files. Returns a source map with verified, conflicting, not-found, and manual-review statuses. Use when the user asks to fact-check a memo, audit, timeline, executive summary, research packet, complaint draft, or project brief against PDFs, XML, HTML, notes, screenshots, or exported chat logs before sending, publishing, or relying on the result. Also use when the user says "check this against the source," "verify the claims," "audit the facts," "does this match the record," or provides a draft alongside source documents. If the user is about to send something important, this skill applies.

Creates distinctive, production-grade frontend interfaces that avoid generic AI aesthetics. Use when the user asks to build web components, pages, artifacts, dashboards, React components, HTML/CSS layouts, landing pages, posters, or any web UI. Also use when the user says "build me a page," "make this look better," "redesign this," "style this," "create a component," or uploads a screenshot and expects a rebuild. If the task produces HTML, CSS, JSX, or any visual web output, this skill applies. Generates creative, polished code with exceptional attention to aesthetic detail. Never produces generic output.

Designs and refines high-performance prompts, system instructions, agent workflows, and multi-step operating guidance for LLMs. Use when the user asks for a system prompt, agent specification, orchestration plan, instruction hierarchy, subagent workflow, prompt debugging, skill file creation, CLAUDE.md writing, or a cleaner prompt architecture. Also use when the user says "write a system prompt," "design an agent," "fix this prompt," "make this more reliable," "build a skill," "orchestrate these agents," or describes any multi-agent or multi-step LLM workflow. If the output is instructions that an LLM will execute, this skill applies.

Turns large document sets into matched multi-volume series with consistent structure, formatting, handoff pages, abbreviations, and output artifacts. Use when the user asks to organize records, reports, or mixed project documents into separate volumes, normalize Markdown/DOCX/PDF packets, create executive and full summary pairs, clean folder layout, or keep multiple document packets stylistically aligned. Also use when the user says "organize these files," "build a series," "normalize these documents," "clean up the folder," "package these records," or has more than 5 documents that need consistent structure. If the task involves multiple documents that must look like they belong together, this skill applies.

Drafts structured regulatory complaints, demand letters, and multi-recipient filing packages with jurisdiction-appropriate citation and first-person assertive voice. Use when the user needs to draft, revise, or format complaints to any government agency, regulatory body, or institutional grievance channel. Covers ADA Title II/III complaints, HIPAA/OCR filings, state medical board complaints, Ohio pre-suit notices, demand letters, FDCPA disputes, CFPB complaints, FTA complaints, and multi-agency escalation packages. Also use when the user says "file a complaint," "draft a demand letter," "write to the agency," "escalate this," or references any regulatory body by name. If the user describes being wronged by an institution, this skill applies.

Executes scoped research and returns decision-ready briefs with source-backed findings, comparisons, open questions, and next actions. Use when the user asks to research a topic, investigate something, compare options, verify current guidance, gather official sources, or turn a broad question into a concise brief. Also use when the user says "look this up," "research this," "what are my options," "find out if," "investigate," "what does the law say about," or asks any question that requires consulting multiple sources before answering. If the answer requires more than one source, this skill applies.

Use when porting a design asset into the JCStream project — a Figma export, JSX mockup, screenshot, design zip, or hand-drawn spec. Translates the input into Jinja templates + CSS tokens + Python helpers while preserving the static-site / no-JS-required contract and accessibility. Trigger phrases: "port this design", "implement the mockup", "translate the Figma", "from this screenshot", "build from this mockup", "convert this JSX", "redesign the homepage from this", "redesign … from this spec", "here's a Figma/PNG/screenshot of the new …", "based on this PNG".

Use when editing user-facing legal language in JCStream templates — presumed-innocent banners, FCRA disclaimer, ORC § 149.43 attribution, ORC § 2953.32 expungement removal protocol, no-fee guarantee, CC BY-NC 4.0 license footer, comment-policy block. Touches web/templates/{index,inmate,stats,statute,data}.html and web/templates/base.html footer (incl. noindex/noarchive meta). Trigger phrases: "update the disclaimer", "rephrase the banner", "fix the FCRA notice", "removal policy", "expungement language", "sealing notice", "takedown protocol", "no-fee guarantee", "license footer", "presumption of innocence".

Use when maintaining the Ohio Revised Code mappings in the JCStream project — data/orc_offenses.json (code → title, degree) and data/explainers.json (code → plain-English explanation). Covers the base-code normalization in scraper/orc.py, the 10-tier degree ladder (F1…MM), and the info-log of unmapped codes. Trigger phrases: "add an ORC code", "explainer for §...", "missing statute title", "fix ORC tier/degree", "add explainer", "ORC code shows ?/unknown", "plain-English for §...", "edit orc_offenses.json or explainers.json".

Use when investigating why a JCStream sweep didn't produce fresh data — the roster count is flat, the changelog is empty, the site rendered stale photos, or the GitHub Actions sweep workflow showed exit-0 but no commit. Covers the silent degraded-roster fallback in scraper/sweep.py, the health thresholds in sweep_guards.py, and how to read changelog.json / history.json / current.json together. Trigger phrases: "sweep didn't update", "stuck count", "no changes in changelog", "scraper looks idle", "stale photos", "partial sweep", "sweep bailed", "sweep timed out", "nameless records", "detail watchdog".

Use when editing or creating Jinja templates under web/templates/ in the JCStream project — base.html, index.html, inmate.html, stats.html, statute.html, data.html, _card.html, feed.xml. Covers css_version cache-busting, data-* filter hooks, body.is-table view toggle, lightbox inert focus management, lightbox/tier-tooltip data-* contracts, and base.html block overrides. Trigger phrases: "edit the inmate page", "add a section to the homepage", "fix template rendering", "edit the masthead/footer", "fix the lightbox", "add a filter", "tweak the table view", "update the RSS template".

Use when writing or updating pytest tests under `tests/` in JCStream — `test_build.py`, `test_sweep.py`, `test_parsers.py`, `test_store.py`, `test_pra_send.py`, etc. Covers offline HTML fixtures in `tests/fixtures/` (DOE/ROE placeholder names, ORC 149.43 no-real-records rule), `monkeypatch`-based network/SMTP mocking, the schema-version guard in `test_store.py`, and the absent `tests/conftest.py`. ≥173 tests must stay green (baseline as of 2026-05-14; suite is expected to grow). Trigger phrases: "write a test for", "add a regression test", "fix the failing test", "add coverage", "test the parser", "fixture for", "the suite is failing", "pytest".