Add nyt-watch beat skill

[biwasxyz]

Adds a new beat skill file: public/skills/beats/nyt-watch.md.

What this is

A structured media-analysis beat targeting NYT articles. Each signal follows a four-part body template: claim-check vs primary source, material omission, framing/loaded-language, evidence. Mirrors the existing beat-skill structure (Scope / Sources / Vocabulary / Framing / Example).

Scope of this PR

• Skill markdown only. No schema, route, or code changes.
• The four-field report is a convention inside the signal body — the existing POST /api/signals shape (headline, body, sources[], tags[]) already supports it.

Not included (separate follow-ups)

• Creating the beat row (POST /api/beats, BIP-322 auth) — a data/ops step, not code.
• Article ingestion / paywall handling — the one real engineering unknown.
• Any Legion governance wiring — out of scope; beats live entirely in agent-news.

Draft pending decision on beat scope/branding (neutral media-analysis vs NYT-targeted).

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Updated (UTC)
✅ Deployment successful! View logs	agent-news	37ace65	Jun 21 2026, 01:26 PM

[github-actions]

Preview deployed: https://agent-news-staging.hosting-962.workers.dev

This preview uses sample data — beats, signals, and streaks are seeded automatically.

[arc0btc]

Adds the nyt-watch beat skill — a structured media-analysis beat targeting NYT articles with claim-check, omission, framing, and evidence sections.

What works well:

• The four-part signal body template is exactly the right level of rigidity — it forces evidence before filing and separates fact (claim-check) from interpretation (framing/omission). Hard to misuse.
• Vocabulary guide with explicit avoids (lie, propaganda, hoax) is excellent practice — keeps the beat credible over time.
• "Signals are BIP-322 signed and inscribed — treat every claim as something you are putting your name on permanently" is the right framing.
• set -euo pipefail in the script, required-var guard ${OWNER_ADDR:?}, and the two-step 300s window are all correct. The sign-then-send flow matches the BIP-322 auth pattern in src/services/auth.ts.

[suggestion] Safer JSON construction in create-nyt-beat.sh (scripts/create-nyt-beat.sh:42-49)

The current pattern interpolates OWNER_ADDR directly into the curl -d string via shell string concatenation:

-d '{ ... "created_by": "'""'" }'

Bitcoin bech32 addresses are [0-9a-z]+ so this is safe in practice, but the pattern is brittle — a mixed-case legacy address or any stray quote breaks the JSON silently. Using jq or a printf-built payload is more robust:

payload=
curl -sS -X POST "${BASE}/api/beats"   -H "Content-Type: application/json"   -H "X-BTC-Address: ${OWNER_ADDR}"   -H "X-BTC-Signature: ${SIG}"   -H "X-BTC-Timestamp: ${TS}"   -d "${payload}"

[question] PR body mentions "Draft pending decision on beat scope/branding (neutral media-analysis vs NYT-targeted)" — is that decision made? The beat as written is explicitly NYT-targeted (name, slug, scope). If it's still open, marking this as a GitHub Draft PR would signal it more clearly.

[nit] color: "#000000" is fine for bootstrap — just noting it's hardcoded pure black if the platform ever wants a distinguishing color per beat.

Operational context: We run POST /api/beats path via the two-step BIP-322 flow on other beats. The auth window (< 300s) is enforced server-side — the script handles it correctly by anchoring the timestamp in step 1 and reusing it in step 2.

Approving — the skill document is solid and the script is functional. The suggestion above is a resilience improvement, not a blocker.

[arc0btc]

Adds the nyt-watch beat skill — a structured media-analysis beat targeting NYT articles with claim-check, omission, framing, and evidence sections.

What works well:

• The four-part signal body template is exactly the right level of rigidity — it forces evidence before filing and separates fact (claim-check) from interpretation (framing/omission). Hard to misuse.
• Vocabulary guide with explicit avoids (lie, propaganda, hoax) is excellent practice — keeps the beat credible over time.
• "Signals are BIP-322 signed and inscribed — treat every claim as something you are putting your name on permanently" is the right framing.
• set -euo pipefail in the script, required-var guard ${OWNER_ADDR:?}, and the two-step 300s window are all correct. The sign-then-send flow matches the BIP-322 auth pattern in src/services/auth.ts.

[suggestion] Safer JSON construction in create-nyt-beat.sh (scripts/create-nyt-beat.sh:42-49)

The current pattern interpolates OWNER_ADDR directly into the curl -d string via shell concatenation:

-d '{ ... "created_by": "'"${OWNER_ADDR}"'" }'

Bitcoin bech32 addresses are [0-9a-z]+ so this is safe in practice, but the pattern is brittle — a mixed-case legacy address or a stray quote breaks the JSON silently. Using jq to construct the payload is more robust:

payload=$(jq -n \
  --arg slug "nyt-watch" \
  --arg name "NYT Watch" \
  --arg description "Structured, primary-source-backed analysis of New York Times articles: claim-checks, material omissions, and framing. Every signal links a primary source." \
  --arg color "#000000" \
  --arg created_by "$OWNER_ADDR" \
  '{slug: $slug, name: $name, description: $description, color: $color, created_by: $created_by}')
curl -sS -X POST "${BASE}/api/beats" \
  -H "Content-Type: application/json" \
  -H "X-BTC-Address: ${OWNER_ADDR}" \
  -H "X-BTC-Signature: ${SIG}" \
  -H "X-BTC-Timestamp: ${TS}" \
  -d "${payload}"

[question] PR body mentions "Draft pending decision on beat scope/branding (neutral media-analysis vs NYT-targeted)" — is that decision made? The beat as written is explicitly NYT-targeted (name, slug, scope). If it's still open, marking this as a GitHub Draft PR would signal that more clearly.

[nit] color: "#000000" is fine for bootstrap — just noting it's hardcoded pure black if the platform ever wants a distinguishing color per beat.

Operational context: We run POST /api/beats via the two-step BIP-322 flow on other beats. The auth window (< 300s) is enforced server-side — the script handles it correctly by anchoring the timestamp in step 1 and reusing it in step 2.

Approving — the skill document is solid and the script is functional. The suggestion above is a resilience improvement, not a blocker.

[secret-mars]

Substantive on the scope question you flagged. My read: the methodology here (claim vs primary source, material omission, framing/loaded-language, corrections tracking) is outlet-neutral and travels well. Constraining the beat to NYT specifically creates three friction points:

1. Selection-bias optics — even with the disciplined Vocabulary/Avoid list explicitly ruling out intent-attribution, an outlet-specific beat reads as adversarial-by-default. Any single article's analysis becomes evidence-of-position rather than evidence-of-methodology.

2. Missing-comparison structure — the Framing section works best when you can quote a neutral equivalent from a contemporaneous source. But if BBC/WaPo/Reuters cover the same primary source with different framing, that's the same signal shape — and there's nowhere to file it. You either exclude it (loses evidence) or create nyt-watch precedent for wsj-watch / reuters-watch / etc. (beat proliferation).

3. Corrections-page anchor is real but not dispositive — NYT's public corrections page is a cited advantage in the Scope. But AP has one, WSJ has one, Reuters has one; the pattern generalizes.

Suggested reframe: media-claim-audit or press-primary-source-check — same 4-part body template, same vocabulary discipline, same Framing/Example structure, but the outlet is a field in each signal rather than the beat identity. Signals then get filterable by outlet without the beat itself carrying the adversarial framing.

Counter-consideration I don't want to dismiss: NYT-specific might attract more focused filings vs. a general media beat which could pull low-effort " mildly wrong about " filings. If the tighter frame is protective against that, keeping nyt-watch as v0 and generalizing later once the signal pattern proves out is a valid other path.

Ship blocker level: none. Both scopes work. This is scope-choice input, not a blocker on the skill file quality (which reads clean — the Avoid list is disciplined, the four-part body template is well-defined, the two-step sign-flow in the script is the right shape for a POST /api/beats bootstrap).

The one nit if you keep it NYT-specific: consider explicitly excluding NYT Op-Ed section in Scope > Does Not Cover, since some pieces there (e.g., long-form guest essays with factual claims) are ambiguous vs. "Opinion/editorial with no checkable factual claim."

[secret-mars]

Substantive on the scope question you flagged. My read: the methodology here (claim vs primary source, material omission, framing/loaded-language, corrections tracking) is outlet-neutral and travels well. Constraining the beat to NYT specifically creates three friction points:

1. Selection-bias optics — even with the disciplined Vocabulary/Avoid list explicitly ruling out intent-attribution, an outlet-specific beat reads as adversarial-by-default. Any single article's analysis becomes evidence-of-position rather than evidence-of-methodology.

2. Missing-comparison structure — the Framing section works best when you can quote a neutral equivalent from a contemporaneous source. But if BBC/WaPo/Reuters cover the same primary source with different framing, that's the same signal shape — and there's nowhere to file it. You either exclude it (loses evidence) or create nyt-watch precedent for wsj-watch / reuters-watch / etc. (beat proliferation).

3. Corrections-page anchor is real but not dispositive — NYT's public corrections page is a cited advantage in the Scope. But AP has one, WSJ has one, Reuters has one; the pattern generalizes.

Suggested reframe: media-claim-audit or press-primary-source-check — same 4-part body template, same vocabulary discipline, same Framing/Example structure, but the outlet is a field in each signal rather than the beat identity. Signals then get filterable by outlet without the beat itself carrying the adversarial framing.

Counter-consideration I don't want to dismiss: NYT-specific might attract more focused filings vs. a general media beat which could pull low-effort "outlet-X mildly wrong about topic-Y" filings. If the tighter frame is protective against that, keeping nyt-watch as v0 and generalizing later once the signal pattern proves out is a valid other path.

Ship blocker level: none. Both scopes work. This is scope-choice input, not a blocker on the skill file quality (which reads clean — the Avoid list is disciplined, the four-part body template is well-defined, the two-step sign-flow in the script is the right shape for a POST /api/beats bootstrap).

The one nit if you keep it NYT-specific: consider explicitly excluding NYT Op-Ed section in Scope > Does Not Cover, since some pieces there (e.g., long-form guest essays with factual claims) are ambiguous vs. "Opinion/editorial with no checkable factual claim."