Haro

/haro — Daily journalist-query response generator

Captures the highest-velocity backlink-acquisition opportunity: journalists actively requesting sources for stories that will publish in the next 3–14 days. A single response landing in a Forbes / WSJ / industry-outlet quote earns one DA-80+ editorial backlink — equivalent to ~6 months of organic-link attrition for a small site.

Important context: the original "HARO" (Help A Reporter Out) was shut down by Cision in December 2024 after a brief rebrand to "Connectively". The skill name /haro is retained per the v2 spec (familiar shorthand for the category), but the underlying source is configurable per site. As of 2026, the active sources are:

Each site picks 1–3 sources via site-info.md → "Journalist sources" block.

Inputs

Required (asked at start if not given):

• Site — must match a folder under sites/[name]/.

Optional (per-run overrides):

• --source=X — pull from only one source this run (default: all configured).
• --min-match-score=N — minimum match score to draft a response (default: 70 out of 100).
• --max-drafts=N — cap responses per run (default: 5).

Workflow

Step 0 — Cron-mode detection (if invoked with --cron)

Cron mode is the DEFAULT pattern for this skill — it's designed for daily firing. Required behavior:

1. Idempotency check: if sites/{site}/_inbox/haro-responses-{date}.md exists for today, exit with idempotent-skip.
2. Escalation: any error pulling a source (API down, auth expired, rate-limit, source ToS change) → write _inbox/haro-source-error-{source}-{date}.md with full error context, continue with remaining sources.
3. Audit log: standard appendRun({ exit, ... }) per scripts/lib/audit-log.mjs.
4. Backoff: at start, checkBackoff({ routine: 'haro-daily' }). If true → pause + escalate per standard contract.

Step 1 — Read context (Rule 1)

Load in order:

1. CLAUDE.md
2. sites/{site}/site-info.md — especially "Journalist sources" block, "Authority" block (what site is on-the-record about), "Person profile" if applicable
3. sites/{site}/references/stats.md — for the numbers that make a quote citable
4. sites/{site}/references/stories.md — for relevant anecdotes
5. sites/{site}/references/voice.md — for tone consistency
6. sites/{site}/backlinks.md — for "Banned outreach targets" + outreach history

If any required file is missing → escalate per cron contract.

Step 2 — Source authentication check

For each source listed in site-info.md → "Journalist sources":

• Verify the corresponding API key / session cookie is present in .env (cron) or available via auth flow (interactive).
• If missing → escalate to _inbox/haro-auth-missing-{source}.md with setup instructions. Skip that source for this run.

Supported auth methods:

• Featured.com: API key in .env as FEATURED_API_KEY. Get from featured.com → Settings → API.
• Source of Sources: email-digest parsing (forward to [site-specific]+sos@yourforwarder.com; cron polls IMAP).
• Qwoted: session cookie scraped via Chrome MCP (Qwoted blocks API access on lower tiers).
• Help a B2B Writer: email-digest parsing (same pattern as Source of Sources).

Step 3 — Pull today's queries

For each authenticated source, fetch queries submitted in the last 24 hours. Normalize to a common schema:

- query_id: featured-2026-05-17-abc123
  source: featured
  outlet: Forbes
  reporter: jane.doe@forbes.com
  reporter_handle: "@jane_doe"
  category: "Business / AI"
  topic: "Examples of AI ROI in SMBs"
  query_body: "[full text of the journalist's request]"
  word_count_requested: 150
  deadline: 2026-05-19T17:00:00Z
  embargo: null
  preferred_credentials: "founders, AI consultants, SMB ops leaders"
  pull_time: 2026-05-17T09:00:00Z

Step 4 — Match-score each query against site expertise

For each query, compute a 0–100 match score from:

Threshold (default --min-match-score=70): only queries scoring ≥ 70 generate a draft response. Lower-scoring queries are logged in the daily summary but no draft is generated.

Step 5 — Draft response for high-match queries

For each query above threshold, generate a draft pitch:

Format: 100–200 words (or word_count_requested ± 20%), structured as:

1. Lead with a number from stats.md — never round, never invent.
2. One concrete example from stories.md (anonymized unless consented) — 1–2 sentences.
3. One contrarian or specific opinion backed by the stat — the unique angle that makes the quote citable.
4. Byline matching the site's documented byline policy (per memory: site-a → "Site A", site-b → co-byline OK, site-c → real founder name).
5. Contact line with the reporter's preferred contact method.

Voice constraints: universal rules from references/voice.md apply — banned words, no exclamation marks, no AI tells. Pitch must read like a real expert wrote it, not like an AI. Run the anti-AI checklist before saving.

No links in the pitch itself — reporters explicitly reject pitches containing URLs as spam. The link gets included only IF the reporter follows up requesting more context.

Step 6 — Quality refusals

Refuse to draft a pitch when:

• Query asks for credentials the site doesn't have (e.g., reporter wants "PhD in computational biology" and site has no such person).
• Query requires fabrication (the answer can't be sourced from stats.md or stories.md).
• Outlet is in backlinks.md "Banned outreach targets" list.
• Reporter has previously declined a pitch from this site (cross-check backlinks.md "Outreach response tracking").
• Query embargo or deadline already passed.

Refused queries are logged with reason in the daily summary; no draft generated.

Step 7 — Output

Write to sites/{site}/_inbox/haro-responses-{date}.md:

# HARO daily responses — {site} — {date}

Pulled {N} queries across {M} sources. Generated {K} drafts ({K2} refused).

## Drafts awaiting your approval

### 1. Forbes — Jane Doe — "Examples of AI ROI in SMBs"
**Match score:** 82
**Deadline:** 2026-05-19 17:00 UTC (48h)
**Outlet DA:** 95
**Source:** Featured.com

**Draft pitch:**
> [pitch body, 150 words]

**To send:** copy the draft into Featured.com query response field. Reply to this inbox item with `sent` to log the outreach. The skill will create the outreach pipeline row in `backlinks.md`.

---

### 2. ...

## Queries refused (logged for transparency)

- "Topic X" — Forbes — refused: no PhD-qualified author available (matching skipped)
- "Topic Y" — Outlet Z — refused: outlet in banned-list (notes.md: prior bad-faith interaction 2025-09-04)

## Queries below match threshold (≥ 50, < 70)

(no draft generated; listed for transparency)

- "Topic A" — Outlet B — score 64 (`stats.md` has no number to anchor a 150-word pitch)
- ...

Step 8 — Update outreach pipeline (on user approval — separate skill)

The cron only DRAFTS. When user replies sent to an inbox item, a follow-up skill (/outreach-log) appends a row to backlinks.md → "Outreach pipeline" with status sent and the 14-day follow-up trigger.

Step 9 — Audit log

Standard appendRun({ exit: 'shipped' | 'idempotent-skip' | 'escalated', metadata: { queries_pulled: N, drafts_generated: K, refusals: K2 } }).

Refusal conditions

This skill refuses to run if:

• Site has no "Journalist sources" block in site-info.md.
• Site has no references/stats.md or references/stories.md (no anchor for citable pitches).
• All configured sources are unauthenticated.

Setup checklist (one-time per site)

Before this skill can run on a new site:

1. Add to site-info.md:

   ## Journalist sources
   
   - **Featured.com:** {profile_url}, API key in .env as FEATURED_API_KEY_{SITE}
   - **Source of Sources:** subscribed to digest at {forwarder_email}
   - **Authority block:** [3–5 areas this site is on-the-record about]
   - **Preferred byline format:** {per byline policy memory}
   

2. Add _inbox/ to site folder if not present.
3. Add cron entry to .claude/schedules.yml (portfolio-daily-haro).
4. Add API keys to .env + trigger prompt (per remote-cron-no-dotenv-access memory — cron secrets must be in trigger prompt, not .env).

Routine versioning + idempotency contract

Idempotency key: (site, date). The skill runs once per site per day. Re-running on the same date skips with idempotent-skip and logs the score snapshot.

Routine version: 1.0.0 (2026-05-17). Increment on schema changes to haro-responses-{date}.md output format.