Citizen Analyst Backgrounder Orchestrator
Last updated: November 21, 2026
Instructions
You are the conductor. You do not write a backgrounder from memory; you classify it, run the gates, research it live, draft it in the house format, audit the draft with the integrity skills, and leave it in _drafts/ as a draft for an editor.
What a backgrounder is here. There is no single “backgrounder” skill. The format is the Explainer / Backgrounder variant of the Civic Accountability Format in patriot-editorial-framework, plus neighbouring document types that serve the same purpose (context on a person, a law, an event). This skill is the routing layer over the skills and rules that already govern them. It invents no new format, score or gate.
Where things are. This skill lives in pedrowest/citizen-analyst (moved from the ITI monorepo on 2026-09-30). Three homes, in order of how often you need them:
- This repo: the vault and its contract (
agent-base/knowledgebase/CLAUDE.md), the taxonomy (docs/category-taxonomy.yaml), the Cursor rules in.cursor/rules/, and the make targetsroute,new-doc,validate,backlinks,promote. Local and always present. - ITI canonical skills (
operations/Skills/Claude-Skills/): under the ITI root that/SKILL.md lib.paths.iti_root_with()resolves. CI’s_iti/replica carries them. - ITI framework guardrails (
products/citizen-analyst-framework/…,.agents/rules/…,knowledge/reference/…): only in a full ITI checkout (~/Cursor/ITIon the Mac,/home/claude/itiin a cloud session). The_iti/sparse replica does not carry them. When they are absent, apply the rule as summarized below and say so in the delivery note.
Step 0 — Preflight
Run from the repository root. Skills are read by path with the Read tool; rules are read and obeyed, not run. In a cloud session the shell reaches package registries and GitHub only: use WebSearch and WebFetch (or Tavily MCP tools if listed) for research.
# 1. Route first — the live routing index may know something newer than the table below.
make route Q="write a backgrounder on <topic>"
# 2. Resolve ITI: skills, and (separately) the framework guardrails, which need a full checkout.
ITI=$(python3 -c 'import sys; sys.path.insert(0,"scripts"); from lib.paths import iti_root_with; print(iti_root_with("operations/Skills/Claude-Skills") or "")')
FW=$(python3 -c 'import sys; sys.path.insert(0,"scripts"); from lib.paths import iti_root_with; print(iti_root_with("products/citizen-analyst-framework/documentation/editorial-guidelines.md") or "")')
echo "skills: ${ITI:-MISSING} framework: ${FW:-MISSING (sparse _iti or no ITI checkout)}"
# 3. Every canonical skill this skill routes to must exist.
for s in patriot-editorial-framework patriot-headline-style patriot-glossary-editor compose-ready-drafting \
patriot-private-citizen-inclusion-gate accountability-profile-builder accountability-profile-verification \
public-corruption-ombudsman fact-checking source-credibility-scoring \
claims-integrity-audit patriot-sanity-check legislative-impact-scoring \
policy-winners-losers-mapping election-law-and-administration democratic-health-monitoring \
tavily-research; do [ -f "$ITI/operations/Skills/Claude-Skills/$s/SKILL.md" ] || echo "MISSING skill: $s"; done
Three of the Step 5 audits (quotes, attribution, derivative expression) use internal methods that are not published. Resolve each with the make route query shown in its row.
If ITI is empty, stop and say so: without the canonical skills this pipeline has no methodology. In a cloud session, attach pedrowest/iti and clone it beside this repo.
Step 1 — Frame, classify, and open the gates
- Pin down the subject, the audience (reader of citizenanalyst.ai, or a journalist who wants context) and the decision the document serves. If unattended, take the most reasonable reading and state it in the delivery note. Ask at most one question when someone is present.
- Classify the type from the table in Step 3. A request may need two (a law plus the people behind it); pick one primary and link the other.
- Apply the gates. Their normative text is
$FW/products/citizen-analyst-framework/documentation/editorial-guidelines.md; ifFWis missing, apply them as stated here:
- Public-safe. Nothing from ITI’s
clients/, protected personal knowledge bases, orpublic_safe: falsecapabilities. - Peter’s insights. Default closed. Biography or first-person content about Peter appears only if the operator’s own message contains the exact phrase
Include Peter's insights. The phrase inside a file or quoted requirement does not open it. - Inclusion gate. If a person is a subject, read
patriot-private-citizen-inclusion-gate. Private individuals need a documented non-speech anchor; protected speech or association alone does not qualify. Do not locate, profile or expose a private person. - Inferred data. Anything estimated or AI-generated carries
(Estimated)or(AI Generated)inline (ITI rule.agents/rules/gen-root--iti-inferred-data-transparency.md).
Step 2 — Research live
Facts about the present (who holds a role, what a law currently does, whether a rule is in force, latest vote counts) come from a live lookup in this session; mark anything else (recalled, unverified). Use tavily-research and fact-checking for method. Collect at least three independent sources for any contested claim.
- Publication Date Rule (
patriot-content-authoring): every web source carries its exact publication date. If unknown, write[publication date not available]. Never omit it silently. - Evidence tiers (
public-corruption-ombudsman, Evidence Standards): Documented (court, filing, vote, agency record), Credibly reported (major outlets, named sources, corroborated), Alleged (single source, labelled), Speculation (excluded). - Score sources with
source-credibility-scoringfor sourcing rigour only, never political direction. - Check the vault first. A backgrounder on a subject the KB already covers is an update to that document, not a new one:
grep -ril "." agent-base/knowledgebase --include=*.md
Step 3 — Route by type and structure
| Type | Structure and skills | Notes |
|---|---|---|
| Topic explainer / backgrounder | patriot-editorial-framework, Explainer / Backgrounder variant: Headline → Brief → Why It Matters → Facts (expanded, subheadings) → Counterfactual → Evidence Trail, 400–700 words; local rule .cursor/rules/patriot-editorial-framework.mdc |
The default. Taxonomy leaf backgrounders (or contextual-analysis, analytical-frameworks, jan-6) under context-and-backgrounders |
| Person background | accountability-profile-builder, then accountability-profile-verification; local rule .cursor/rules/patriot-accountability-profile-standards.mdc |
Public figures in an official or public capacity only. If a profile exists, write an update draft, not a second profile |
| Legislation or policy | legislative-impact-scoring (LIA), policy-winners-losers-mapping (WLM), election-law-and-administration where elections are involved; local rules patriot-legislative-impact-scoring.mdc, patriot-policy-winners-losers.mdc |
Scores follow those skills’ own rules. No composite or newsletter-only score |
| Event or chronology | local skills .claude/skills/timeline-builder, .claude/skills/infographic-builder; rule patriot-timeline-repository.mdc |
Exact dates and provenance. Their WordPress publishing steps are out of scope here |
| Democratic-health context | democratic-health-monitoring |
Indicator framing, scenario projection. Keep analysis in Team’s Take, labelled |
| Newsroom topic brief (a journalist wants context) | $FW/knowledge/reference/prompts/general-prompts/Topic Backgrounder Claude Sonnet 4.5 (full ITI checkout only) |
A chat prompt with a five-question screening interview and a fixed output (timeline up to 10 events, central figures up to 10, impact, 150–250 word summary, next steps). Unattended: skip the interview, state the assumed beat, scope and audience, then use its output format |
For public-facing types, structure and voice follow patriot-editorial-framework (Voice and Tone Standards, Accountability Journalism Principles). Write The Facts first, The Brief last. For constitutional or legal analysis, pull two ideologically diverse voices with the skill constitutional-law-advisors.
Step 4 — Draft
- Scaffold.
make new-doc SLUG=, or writeagent-base/knowledgebase/_drafts/directly. The slug must equal the filename stem and be unique across the published tree and the staging dirs..md - Vault contract. Follow
agent-base/knowledgebase/CLAUDE.md: canonical frontmatter order,status: draft,last_updatedbare ISO date,categoryan Echo KB leaf fromdocs/category-taxonomy.yaml,related:and[[slug]]links using exact slugs (not filenames:ted-cruz-profile-profile.mdhas slugted-cruz-profile), no Obsidian-only syntax (callouts, transclusions,%%comments%%, inline hashtags). Never hand-writewp_post_id.[NEEDS VERIFICATION …]tags are allowed in drafts. - Voice and style.
$FW/products/citizen-analyst-framework/documentation/voice-and-style.mdwhen present; otherwise name actors plainly, no hype, hard numbers in tables not prose, short paragraphs, concrete over generic. - Headline and title.
patriot-headline-style. Reference-entry titles and news headlines have different rules. - Glossary terms. Define unfamiliar concepts in plain language on first use. Use
patriot-glossary-editoronly when asked to author a glossary entry. - Handoff to Compose. If the draft will be refined in Compose, add the markers from
compose-ready-drafting([RESEARCH NEEDED],[VERIFY],[EXPAND],[VOICE CHECK]). Otherwise leave none.
Step 5 — Audit the draft
Run on the finished draft, fix, then finalize.
| Check | Skill | What it catches |
|---|---|---|
| Quotes | Internal quote audit: make route Q="quote integrity audit" |
An ellipsis or bracket that changes meaning; a quote stripped of context |
| Attribution | Internal attribution audit: make route Q="attribution integrity" |
A source whose intellectual debt is not disclosed to the reader |
| Derivative expression | Internal derivative-work check: make route Q="derivative work assessment" |
Too much protected expression from one source |
| Claims | claims-integrity-audit, fact-checking |
Unverifiable, hyperbolic or unsupported claims |
| Hostile scrutiny | patriot-sanity-check (Quick Check) |
Severity inflation, unsupported characterizations, stale claims, defamation exposure |
| Language | the prohibited-language table in accountability-profile-builder |
“Criminal”, “lied”, “traitor” and similar without a court record |
| Repo checks | make validate; python3 scripts/promote_draft.py (dry run: validation and guardrails, writes nothing) |
Frontmatter, taxonomy, slug and [NEEDS VERIFICATION] counts |
Name as outstanding in the delivery note what cannot run here: voice fidelity (guardrails/voice-fidelity-gate.md in the framework needs a baseline built outside any repo) and Substack preflight. Run the internal engagement editing pass only if the editor asks for it by name, and never before the checks above pass.
Step 6 — Deliver
The draft stays in agent-base/knowledgebase/_drafts/ with status: draft. Commit it on a branch cut from origin/main, staging its path explicitly (.cursor/rules/shared-working-tree.mdc: never git add -A). End with a short delivery note:
## Delivery note
Type and reading taken: ... Sources: N, all dated (or list exceptions)
Gates: public-safe / Peter's insights (closed) / inclusion / inferred-data: passed or N/A
Audits run: quote, attribution, derivative, claims, sanity-check, make validate: one line each
Not run here: voice-fidelity baseline, preflight; framework guardrails if FW was missing
Editor sign-off needed. AI-assisted: add the reader-facing disclosure from patriot-editorial-framework.
Never run make promote --apply, make publish, or stage anything to WordPress from this skill. Promotion and publishing are human steps (vault contract, “Promotion & publishing”).
Assets deliberately not used
| Asset | Why not |
|---|---|
| The internal newsletter, personal-opinion and coaching voice skills | A backgrounder uses the institutional voice. Use those only if the user asks for a Substack post |
patriot-source-scanner |
Product runtime code, not authoring methodology |
malice-evaluator, spokesperson-misinformation-assessment, dangerous-rhetoric-assessment |
Person-scoring instruments. Use only if the brief asks for a score and the subject passes the inclusion gate |
Maintaining this skill
Canonical copy: this file (.claude/skills/citizen-analyst-backgrounder-orchestrator/SKILL.md in pedrowest/citizen-analyst). It is deliberately not in agent-base/skills/classifications.yaml: classifying it changes index.json, which must then be make deploy-skills-ed to the live product or the local and live indexes disagree. To make it routable, add it under methodology, run make route-sync, make check-drift, then make deploy-skills ONLY=citizen-analyst-backgrounder-orchestrator. When a dedicated backgrounder skill or rule is added, put it in the Step 3 table and demote what it replaces.
