feat(skills): anti-fabrication gate + real DataForSEO intel for mkt-competitive-brief

[mt-alarcon]

What

Upgrades the existing mkt-competitive-brief skill (previously SKILL.md only) with a verifiability-first workflow and a real data collector, plus the shared DataForSEO client it depends on.

Why

Competitive analysis is one of the easiest places for an LLM to present its training-data memory as fact. This change makes the skill provenance-gated and gives it a path to real, sourced data instead of plausible-sounding guesses.

Changes

SKILL.md — adds a mandatory provenance gate: every competitor data point needs source + URL + collection date, or it lands in a provenance table as unverified / research. Adds source-limit disclaimers up front so the skill never overpromises:

• Meta Ad Library API is restricted to political/social-issue + EU-only in 2026 — commercial creatives are UI-only, no public commercial API.
• Google Ads Transparency Center has no public API (no spend / keyword-trigger / targeting). UI-only.
• Competitor spend and ad keyword-triggers are not observable in any public source — any number is a third-party estimate (label the 15-30% error band).

scripts/competitor_intel.py — real competitive intel via DataForSEO:

1. SERP features + AI Overview presence + intent per keyword
2. Keyword gap (domain-vs-domain) via DataForSEO Labs domain_intersection

Every output carries source + url + collected_at. No credentials → status N/A with instructions (never fabricates, never breaks the pipeline). Reuses the mkt-seo-ops DataForSEO client via importlib instead of duplicating the HTTP client / credentials.

.claude/skills/mkt-seo-ops/dataforseo_client.py — the shared DataForSEO client that competitor_intel.py imports (HTTP Basic, env-based creds, graceful degradation). Included so the new collector is functional out of the box.

tests/ — 38 tests, all network blocked (autouse urllib guard), covering N/A paths, provenance fields, location validation (no silent BR default), truncation, and gap parsing.

Config

Credentials are env-only and both optional: DATAFORSEO_LOGIN / DATAFORSEO_PASSWORD. The skill works without them (provenance-gated manual research); the collector just reports N/A.

Test

$ python3 -m pytest tests/ -q
38 passed
$ python3 scripts/competitor_intel.py --check   # no creds
{"client_loaded": true, "available": false, "reason": null}

🤖 Generated with Claude Code

[sourcery-ai]

Sorry @mt-alarcon, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery