gkit v0.3.0: operation journal, Bing search-health writes, PageSpeed

README.md

@@ -4,7 +4,7 @@ Profile-bound CLI for agent-first access to growth providers. This repository
 has one CLI and one workspace package: `gkit`.
 
 The reviewed provider surface includes DataForSEO, PostHog, Google Ads, Google
-Search Console, Bing Webmaster, and HubSpot.
+Search Console, Bing Webmaster, HubSpot, and PageSpeed Insights.
 
 ## Requirements
 
@@ -107,6 +107,11 @@ Provider execution must bind exactly one App profile. Create one JSON file at
       "secrets": {
         "accessToken": "env:MY_APP_HUBSPOT_ACCESS_TOKEN"
       }
+    },
+    "pagespeed": {
+      "secrets": {
+        "apiKey": "env:MY_APP_PAGESPEED_API_KEY"
+      }
     }
   }
 }
@@ -157,7 +162,7 @@ reviewing the request:
 
 ```bash
 gkit --profile my-app gsc api call \
-  --operation-id gsc.properties.list \
+  --capability-id gsc.properties.list \
   --input '{}' \
   --out ./gsc-properties-plan.json \
   --dry-run
@@ -167,7 +172,7 @@ Remove `--dry-run` only when the profile and request are correct:
 
 ```bash
 gkit --profile my-app gsc api call \
-  --operation-id gsc.properties.list \
+  --capability-id gsc.properties.list \
   --input '{}' \
   --out ./gsc-properties.json
 ```
@@ -181,7 +186,7 @@ it gkit fails with `EFFECT_NOT_ALLOWED` before any request:
 
 ```bash
 gkit --profile my-app gsc api call \
-  --operation-id gsc.sitemaps.submit \
+  --capability-id gsc.sitemaps.submit \
   --input '{"feedpath":"https://example.com/sitemap.xml"}' \
   --allow-write \
   --out ./gsc-sitemap-submit.json
@@ -192,6 +197,11 @@ Search Console; a Restricted user receives `AUTH_FAILED`. Google returns an
 empty body, so the artifact is a gkit receipt with `siteUrl`, `feedpath`, and
 `httpStatus`.
 
+Each live call is journaled under `meta.operationId`. Run `gkit operations` to
+see recent operations and anything still `unknown`; reconcile those with
+`gkit operations reconcile` and a provider evidence reference instead of
+rerunning them.
+
 ### HubSpot read-only example
 
 HubSpot uses one profile-bound private-app access token and calls the REST API
@@ -205,7 +215,7 @@ cannot be dispatched.
 gkit --profile my-app hubspot doctor
 gkit describe --id hubspot.crm.objects.search
 gkit --profile my-app hubspot api call \
-  --operation-id hubspot.crm.objects.search \
+  --capability-id hubspot.crm.objects.search \
   --input @hubspot-search.json \
   --out ./hubspot-contact-search.json \
   --dry-run
@@ -221,6 +231,21 @@ for the analysis. Search pages are capped at 200 and one search query cannot
 page beyond 10,000 results; other list surfaces use lower reviewed page and
 total-result bounds documented by `describe`.
 
+### PageSpeed Insights example
+
+`pagespeed.run` runs Lighthouse for one URL and returns category scores plus
+whether Chrome UX Report field data exists for the URL and its origin; the raw
+PageSpeed response is the artifact. The `apiKey` secret is optional: without it
+requests share Google's low anonymous quota. Runs take tens of seconds.
+
+```bash
+gkit --profile my-app pagespeed doctor
+gkit --profile my-app pagespeed api call \
+  --capability-id pagespeed.run \
+  --input '{"url":"https://example.com/","strategy":"mobile","categories":["performance","seo"]}' \
+  --out ./pagespeed-home.json
+```
+
 ## Configure an Agent
 
 Agents do not need provider-specific CLIs or their own copies of credentials.

docs/adr/0027-journal-every-live-operation-from-the-provider-seam.md

@@ -0,0 +1,10 @@
+---
+type: ADR
+status: accepted
+---
+
+# Journal every live Operation from the Provider Module seam
+
+The Operation Journal (`operations.jsonl`, ADR-0015) now records every live Growth Operation for every provider and effect, and the DataForSEO-only spend ledger is removed. An Operation enters the Journal with one `attempt_started` event written immediately before the Provider Module receives control (ADR-0009), settles with `attempt_settled`, and links its published Evidence with `artifact_recorded`. Spend is not a separate store: a spend capability's `attempt_started` carries the reviewed spend authorization, and every settlement of that Operation carries its spend facts, because the authorization and settlement are created, settled, and reconciled together with the Operation. Spend blockers, Operation State (`unknown`, `not_dispatched`, `succeeded`, `failed`), and `succeeded`-requires-Evidence (ADR-0012) are projections over these events, never stored statuses.
+
+This supersedes the start point of ADR-0003 and the vocabulary of ADR-0014. Calls that fail syntax, validation, profile, effect, or credential checks, and dry-runs, have no provider effect and are not journaled. A settlement whose provider result or spend amount is unknown stays open and changes only through an evidence-backed manual reconciliation of that same Operation (ADR-0013); a confirmed provider result is never changed. `--capability-id` now selects the capability and `meta.operationId` carries the runtime identity, completing ADR-0008. No migration reads the removed `ledger.jsonl`; no host had one when this landed.