Questions.
Straight answers.
What is IMPACT?
Organic operations pack. Decision first, calendar second.
Grok or Claude?
Both. Same folders.
Do I run 33 skills a day?
No. Run impact. The rest force a lane or print a report.
Search Console?
MCP or CSV. Not required to start. Required for click work.
Does it publish?
No.
Does it send outreach?
No.
Will ChatGPT name us?
IMPACT can rewrite a passage. It does not promise a cite.
Nothing worth doing?
Record "nothing beats the bar." That is a successful close.
Where is state?
One vault: .seo/ in the site repo, or artifacts/impact/<slug>/.
How do I install it?
Unzip. Copy every folder that contains SKILL.md into ~/.grok/skills/ or ~/.claude/skills/. Open a new thread.
Why "every folder" — not just impact/?
Command skills are separate folders. The model discovers them by description, not by you pointing at the zip.
Old thread does not see the pack. Now what?
Open a new thread. Old threads do not reload skill folders.
What is the first command?
Run IMPACT on example.com — with the real host. Name the domain every new thread. State is per domain.
What happens on a run?
Vault check, probe of public pages, scoring, one action (or nothing beats the bar), and a run record plus a ticket.
How many actions per run?
One. That is the point.
What is a run record?
The dated row in runs/ naming what was chosen and why. Evidence lives next to the decision.
Does it store my Google credentials?
No. mcp-gsc connects on your client. IMPACT reads the numbers; it does not keep your secrets.
Can I start without Search Console?
Yes. What it cannot see is labeled unknown — required, not optional. Click work wants real data.
How do I feed Search Console offline?
Export Pages (and Queries) from GSC into imports/, then gsc_ingest.py against the prior file.
What is the vault?
One folder per client: config.json, memory.md, KPIs, the action log, and dated tickets.
Where does the vault live?
In order: the vault_path in config.json, else <site-repo>/.seo/, else artifacts/impact/<slug>/.
Can I keep two vaults for one domain?
No. Run upgrade_state.py and delete the empty one.
What is a money URL?
A page that converts. You set 5–15 of them in the watchlist so decay shows up instead of vibes.
What is the scoring formula?
(impact × confidence × learn) / cost. Clears the bar when score ≥ 2.0 and confidence ≥ 0.35.
What counts as impact?
5 blocked money URL or false claim; 4 snippet or template speed on a converting URL; 3 refresh with demand; 2 new page for a winnable intent; 1 hygiene; 0 speculative.
Why does learn exist?
It tunes the lanes after four scored outcomes in the log. Do not mark hit from vibes — title work stays pending until a later number.
Why did it repair instead of write?
Repair, refresh, and correction beat creation when the protected score is at least 85% of the creation score.
What does nothing beats the bar mean?
A clean negative. Record it. That is a successful close, not a failed run.
What does it never invent?
Rankings, referring domains, Core Web Vitals, revenue, and citations. No invented position 8.
Can it publish to my site?
No production publish, no outreach, no paid links. IndexNow only with a real key and a clear ask.
My CMS is webflow / wordpress / hubspot / next. Then what?
Patches are written in that tool's fields, not as raw HTML. config.cms decides which.
CMS is unknown. Then what?
A patch of HTML <title>, meta, and a 301 row for you to apply.
What is radar.md for?
Demand seeds — jobs to be done. pr/angles.md is stories. Do not merge them.
A script tracebacked. What do I do?
Run --help, then point it at a real URL. If it still crashes, that is a script bug.
It ignored scoring and wrote a blog while a money URL 404s?
Reject it and point at the vault log. The run decided; the run logs why.
Audit vs a run — which do I ask for?
seo-audit prints a reading. Run IMPACT takes one action. Do not treat a 40-row audit as today's execute list.
Something else? david@presstitan.com — or ask in the thread and name the domain.