gravity-cli 0.7.3__tar.gz → 0.7.5__tar.gz

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/PKG-INFO

@@ -1,12 +1,12 @@
 Metadata-Version: 2.5
 Name: gravity-cli
-Version: 0.7.3
+Version: 0.7.5
 Summary: gravity cli — init, run, and push python agents to the Gravity marketplace
 License-Expression: Apache-2.0
 License-File: LICENSE
 License-File: NOTICE
 Requires-Python: >=3.12
-Requires-Dist: gravity-schema==0.7.3
+Requires-Dist: gravity-schema==0.7.5
 Requires-Dist: httpx>=0.27.0
 Requires-Dist: keyring>=25.0.0
 Requires-Dist: python-dotenv>=1.0.0

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/packages/gravity-schema/gravity_schema/catalog.yaml

@@ -65,6 +65,24 @@
   guarantee: certified
   side_effect: read
 
+- name: browser.read
+  slug: null
+  toolkit: null
+  description: read a page of a site the person signed in to once (an account the manifest declares) through Gravity's browser worker, as that person; returns the final url, the page text and its links. Only pages of the account's own site. The agent never sees the login. Charged to the run's budget
+  scopes: []
+  breadth: generic
+  guarantee: certified
+  side_effect: read
+
+- name: browser.act
+  slug: null
+  toolkit: null
+  description: do one typed action as the person on a site account they signed in to (linkedin.com send_dm, connect, comment; x.com send_dm, reply), with the text given; paced and capped per account per day by the gateway. Must be under approvals.required_for. Returns what was done, or a refusal code (NOT_SIGNED_IN, NOT_CONNECTED, DAILY_CAP, BLOCKED_BY_SITE)
+  scopes: []
+  breadth: generic
+  guarantee: certified
+  side_effect: write
+
 - name: supadata.get_transcript
   slug: null
   toolkit: null
@@ -354,6 +372,15 @@
   guarantee: certified
   side_effect: read
 
+- name: apollo.enrich_person
+  slug: APOLLO_PEOPLE_ENRICHMENT
+  toolkit: apollo
+  description: match one person in Apollo from their name and company domain (or LinkedIn URL) and return their work email, title and profile links; uses the person's own Apollo account and its credits
+  scopes: []
+  breadth: generic
+  guarantee: community
+  side_effect: read
+
 - name: hunter.find_email
   slug: HUNTER_EMAIL_FINDER
   toolkit: hunter
@@ -372,6 +399,60 @@
   guarantee: community
   side_effect: read
 
+- name: scrapecreators.linkedin_profile
+  slug: SCRAPECREATORS_GET_LINKEDIN_PROFILE
+  toolkit: scrapecreators
+  description: read one public LinkedIn person profile by URL (experience with descriptions, education, about and recent activity) without the person's or the user's login; on Gravity's account through Composio Instant, 1 credit a profile
+  scopes: []
+  breadth: generic
+  guarantee: community
+  side_effect: read
+
+- name: scrapecreators.linkedin_post
+  slug: SCRAPECREATORS_GET_LINKEDIN_POST
+  toolkit: scrapecreators
+  description: read one public LinkedIn post by URL (text, author, date, engagement and comments); on Gravity's account through Composio Instant
+  scopes: []
+  breadth: generic
+  guarantee: community
+  side_effect: read
+
+- name: scrapecreators.twitter_profile
+  slug: SCRAPECREATORS_GET_TWITTER_PROFILE
+  toolkit: scrapecreators
+  description: read one public X/Twitter profile by handle (bio, links, verification, audience); on Gravity's account through Composio Instant
+  scopes: []
+  breadth: generic
+  guarantee: community
+  side_effect: read
+
+- name: scrapecreators.twitter_tweets
+  slug: SCRAPECREATORS_LIST_TWITTER_USER_TWEETS
+  toolkit: scrapecreators
+  description: read the public tweets X shows for one handle (about 100 popular ones, not a full history); on Gravity's account through Composio Instant
+  scopes: []
+  breadth: generic
+  guarantee: community
+  side_effect: read
+
+- name: scrapecreators.reddit_search
+  slug: SCRAPECREATORS_SEARCH_REDDIT
+  toolkit: scrapecreators
+  description: search Reddit posts or comments by keyword with sort and timeframe, one page per call with a cursor; on Gravity's account through Composio Instant, no Reddit API key needed
+  scopes: []
+  breadth: generic
+  guarantee: community
+  side_effect: read
+
+- name: scrapecreators.social_profiles
+  slug: SCRAPECREATORS_FIND_CREATOR_SOCIAL_PROFILES
+  toolkit: scrapecreators
+  description: find a person's linked profiles on other social platforms from one known profile; on Gravity's account through Composio Instant
+  scopes: []
+  breadth: generic
+  guarantee: community
+  side_effect: read
+
 - name: exa.search
   slug: EXA_SEARCH
   toolkit: exa

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/packages/gravity-schema/gravity_schema/contract.py

@@ -82,6 +82,10 @@ MAX_PREVIOUS_KB: Final[int] = 256
 
 # Declared actions (manifest `actions`).
 MAX_ACTIONS: Final[int] = 20
+# Sites the gateway's browser worker has sign-in and recipes for (accounts, browser.read, browser.act).
+ACCOUNT_SITES: Final[frozenset[str]] = frozenset({"linkedin.com", "x.com"})
+MAX_ACCOUNTS: Final[int] = 5
+BROWSER_TOOLS: Final[frozenset[str]] = frozenset({"browser.read", "browser.act"})
 
 
 # Generous bounds, so a pathological manifest can't tie up the validator.

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/packages/gravity-schema/gravity_schema/errors.py

@@ -39,6 +39,7 @@ class ErrorCode(str, Enum):
     INVALID_CONDITION = "INVALID_CONDITION"
     INVALID_ACTION = "INVALID_ACTION"
     INVALID_MEMORY = "INVALID_MEMORY"
+    INVALID_ACCOUNT = "INVALID_ACCOUNT"
     LOCKFILE_MISMATCH = "LOCKFILE_MISMATCH"
     MISSING_FIELD = "MISSING_FIELD"
     UNKNOWN_FIELD = "UNKNOWN_FIELD"

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/packages/gravity-schema/gravity_schema/manifest.py

@@ -726,6 +726,42 @@ class ActionSpec(BaseModel):
         return self
 
 
+class AccountSpec(BaseModel):
+    """A site account the person signs in to once on a Gravity page (Accounts vault). The agent never sees the
+    login: the gateway's browser worker opens the session for browser.read and browser.act calls only."""
+
+    model_config = _STRICT
+
+    name: Annotated[str, Field(max_length=C.MAX_NAME_LENGTH)]
+    site: str
+    label: Annotated[str, Field(max_length=C.MAX_LABEL_LENGTH)] = ""
+    purpose: Annotated[str, Field(max_length=C.MAX_LABEL_LENGTH)] = ""
+    required: bool = False
+
+    @field_validator("name")
+    @classmethod
+    def _check_name(cls, v: str) -> str:
+        if not C.INPUT_NAME_RE.match(v):
+            raise PydanticCustomError(
+                ErrorCode.INVALID_ACCOUNT.value,
+                "Account name '{value}' is not a valid identifier.",
+                {"value": v, "remediation": "Use lowercase letters, digits and underscores, e.g. linkedin."},
+            )
+        return v
+
+    @field_validator("site")
+    @classmethod
+    def _check_site(cls, v: str) -> str:
+        site = v.strip().lower().removeprefix("www.")
+        if site not in C.ACCOUNT_SITES:
+            raise PydanticCustomError(
+                ErrorCode.INVALID_ACCOUNT.value,
+                "Accounts on '{value}' are not supported.",
+                {"value": v, "remediation": f"Use one of: {', '.join(sorted(C.ACCOUNT_SITES))}."},
+            )
+        return site
+
+
 class OfferAs(str, Enum):
     """How the actions under a result are shown."""
 
@@ -767,6 +803,8 @@ class Manifest(BaseModel):
     actions: Annotated[list[ActionSpec], Field(max_length=C.MAX_ACTIONS)] = Field(default_factory=list)
     offer_as: OfferAs = OfferAs.BUTTONS
     offer_label: Annotated[str, Field(max_length=C.MAX_LABEL_LENGTH)] | None = None
+    # Site accounts the person signs in to once (Accounts vault); browser.* tools act through them.
+    accounts: Annotated[list[AccountSpec], Field(max_length=C.MAX_ACCOUNTS)] = Field(default_factory=list)
 
     @field_validator("contract")
     @classmethod
@@ -897,6 +935,31 @@ class Manifest(BaseModel):
             )
         return self
 
+    @model_validator(mode="after")
+    def _check_accounts(self) -> Manifest:
+        names = [a.name for a in self.accounts]
+        dupes = sorted({n for n in names if names.count(n) > 1})
+        if dupes:
+            raise PydanticCustomError(
+                ErrorCode.INVALID_ACCOUNT.value,
+                "Account name(s) declared twice: {value}.",
+                {"value": ", ".join(dupes), "remediation": "Give each account its own name."},
+            )
+        browser = sorted(self.declared_tools & C.BROWSER_TOOLS)
+        if browser and not self.accounts:
+            raise PydanticCustomError(
+                ErrorCode.INVALID_ACCOUNT.value,
+                "{value} acts through a person's own site account, and none is declared.",
+                {"value": ", ".join(browser), "remediation": "Declare the site under `accounts` (e.g. name: linkedin, site: linkedin.com)."},
+            )
+        if "browser.act" in self.declared_tools and "browser.act" not in self.approvals.required_for:
+            raise PydanticCustomError(
+                ErrorCode.APPROVAL_REQUIRED.value,
+                "browser.act sends messages as the person, so each call needs their approval.",
+                {"remediation": "Add browser.act under approvals.required_for, with a fallback for when it isn't approved."},
+            )
+        return self
+
     @model_validator(mode="after")
     def _check_memory(self) -> Manifest:
         if self.state is not None and self.memory is not None:

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/packages/gravity-schema/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "gravity-schema"
-version = "0.7.3"
+version = "0.7.5"
 description = "Gravity agent manifest contract — models, validation, and JSON Schema export"
 readme = "README.md"
 license = "Apache-2.0"

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/packages/gravity-schema/schema/manifest.schema.json

@@ -1,5 +1,43 @@
 {
   "$defs": {
+    "AccountSpec": {
+      "additionalProperties": false,
+      "description": "A site account the person signs in to once on a Gravity page (Accounts vault). The agent never sees the\nlogin: the gateway's browser worker opens the session for browser.read and browser.act calls only.",
+      "properties": {
+        "label": {
+          "default": "",
+          "maxLength": 200,
+          "title": "Label",
+          "type": "string"
+        },
+        "name": {
+          "maxLength": 64,
+          "title": "Name",
+          "type": "string"
+        },
+        "purpose": {
+          "default": "",
+          "maxLength": 200,
+          "title": "Purpose",
+          "type": "string"
+        },
+        "required": {
+          "default": false,
+          "title": "Required",
+          "type": "boolean"
+        },
+        "site": {
+          "title": "Site",
+          "type": "string"
+        }
+      },
+      "required": [
+        "name",
+        "site"
+      ],
+      "title": "AccountSpec",
+      "type": "object"
+    },
     "ActionSpec": {
       "additionalProperties": false,
       "description": "One thing the agent can do. The runner passes the chosen ones in state[\"actions\"].",
@@ -736,6 +774,14 @@
   "additionalProperties": false,
   "description": "What a builder declares about an agent: identity, entrypoint, consumer inputs, and the tools it is permitted to call. Contract version v1.",
   "properties": {
+    "accounts": {
+      "items": {
+        "$ref": "#/$defs/AccountSpec"
+      },
+      "maxItems": 5,
+      "title": "Accounts",
+      "type": "array"
+    },
     "actions": {
       "items": {
         "$ref": "#/$defs/ActionSpec"

gravity_cli-0.7.5/packages/gravity-schema/tests/test_accounts.py

@@ -0,0 +1,42 @@
+"""Site accounts (the Accounts vault) and the browser tools that act through them."""
+
+import pytest
+from gravity_schema import dev_catalog, load_manifest_text, validate_manifest
+from gravity_schema.errors import ManifestValidationError
+
+BASE = {"name": "leads", "runtime": {"entrypoint": "src/agent.py:graph"}}
+LINKEDIN = {"name": "linkedin", "site": "www.LinkedIn.com", "label": "Your LinkedIn", "purpose": "send messages you approve"}
+
+
+def codes(manifest):
+    with pytest.raises(ManifestValidationError) as e:
+        validate_manifest(manifest)
+    return {i.code.value for i in e.value.issues}
+
+
+def test_accounts_with_browser_tools_validate():
+    m = validate_manifest({**BASE, "accounts": [LINKEDIN, {"name": "x", "site": "x.com"}],
+                           "permissions": {"tools": ["browser.read", "browser.act"]},
+                           "approvals": {"required_for": ["browser.act"]}})
+    assert [a.site for a in m.accounts] == ["linkedin.com", "x.com"] and not m.accounts[0].required
+
+
+def test_account_rules():
+    assert codes({**BASE, "accounts": [{"name": "fb", "site": "facebook.com"}]}) == {"INVALID_ACCOUNT"}
+    assert codes({**BASE, "accounts": [LINKEDIN, LINKEDIN]}) == {"INVALID_ACCOUNT"}
+    assert codes({**BASE, "accounts": [{"name": "Linked In", "site": "linkedin.com"}]}) == {"INVALID_ACCOUNT"}
+    assert codes({**BASE, "permissions": {"tools": ["browser.read"]}}) == {"INVALID_ACCOUNT"}  # no account to act through
+    assert codes({**BASE, "accounts": [LINKEDIN], "permissions": {"tools": ["browser.act"]}}) == {"APPROVAL_REQUIRED"}
+
+
+def test_browser_and_apollo_tools_are_in_the_catalog():
+    names = dev_catalog().names()
+    assert {"browser.read", "browser.act", "apollo.enrich_person"} <= names
+    text = """
+name: leads
+runtime: {entrypoint: src/agent.py:graph}
+accounts: [{name: linkedin, site: linkedin.com}]
+permissions: {tools: [browser.read, browser.act, apollo.enrich_person]}
+approvals: {required_for: [browser.act]}
+"""
+    assert load_manifest_text(text, catalog=dev_catalog()).accounts[0].name == "linkedin"

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/pyproject.toml

@@ -1,13 +1,13 @@
 [project]
 name = "gravity-cli"
-version = "0.7.3"
+version = "0.7.5"
 description = "gravity cli — init, run, and push python agents to the Gravity marketplace"
 readme = "README.md"
 license = "Apache-2.0"
 license-files = ["LICENSE", "NOTICE"]
 requires-python = ">=3.12"
 dependencies = [
-    "gravity-schema==0.7.3",  # released in lockstep: always the same version as this package
+    "gravity-schema==0.7.5",  # released in lockstep: always the same version as this package
     "typer>=0.15.0",
     "httpx>=0.27.0",
     "rich>=13.9.0",

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/src/gravity_cli/CHANGELOG.md

@@ -4,6 +4,15 @@ Shown by `gravity update` and `gravity changelog`, and after an update. Each rel
 then one bullet per change: `- new|improved|fixed: A one-line headline. The detail behind it.`
 Write it for a builder, not for a maintainer. `tests/test_changelog.py` fails a release that has no entry.
 
+## 0.7.5 - 2026-10-24
+
+- new: ScrapeCreators tools on Gravity's account. scrapecreators.linkedin_profile reads a public LinkedIn profile in full (experience with descriptions, education, recent activity) where LinkedIn refuses a plain read, plus linkedin_post, twitter_profile, twitter_tweets, reddit_search (no Reddit API key) and social_profiles. They run through Composio Instant, charged to the run's budget.
+
+## 0.7.4 - 2026-10-24
+
+- new: Site accounts. Declare `accounts` (linkedin.com, x.com) and the person signs in once on Gravity; browser.read reads their pages and browser.act sends a DM, a connection note, a comment or a reply they approve, paced and capped per day. Agent code never sees the login.
+- new: apollo.enrich_person matches a person in Apollo from their name and company domain and returns their work email and profile links, on the person's own Apollo account.
+
 ## 0.7.3 - 2026-10-23
 
 - fixed: gravity push fits bigger agents. The manifest travels in a 12 KB request header and every unset field went up as a null, about 100 bytes per input and action; they are now left out, so an agent with a full form and many actions uploads.

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/src/gravity_cli/docs/Gravity.agent.md

@@ -307,6 +307,9 @@ actions:                           # what the agent can do (section 10); leave o
 offer_as: buttons                  # buttons (default) | question: how the actions show under the result
 offer_label: "What next?"
 
+# accounts:                        # a site account the person signs in to once on Gravity (section 6, browser.*)
+#   - { name: linkedin, site: linkedin.com, label: "Your LinkedIn", purpose: "send messages you approve" }
+
 outputs:                           # what a person may ask the result to become (section 9), at most 8
   summary: {renderer: markdown, label: "Summary", destinations: [download, gmail.draft, gmail.send]}
   report:  {renderer: docx, label: "Report (Word)"}
@@ -349,6 +352,7 @@ Rules the validator enforces beyond field types:
 - **Actions** (`INVALID_ACTION`): names unique; `requires` and `then` name other declared actions and never loop;
   at least one is `default: true`, and a default action's `requires` are defaults declared before it; a
   `run_once` action has no `schedule` and is not a `then` target; an action's own inputs have no action-choice.
+- **Accounts** (`INVALID_ACCOUNT`): sites are linkedin.com or x.com; names unique identifiers; `browser.read` or `browser.act` needs at least one account, and `browser.act` must be under `approvals.required_for` (`APPROVAL_REQUIRED`).
 - **Memory** (`INVALID_MEMORY`): `memory` and `state` are not both declared; collection names are identifiers;
   there is no shared-across-people collection.
 - **Quote** `python: "3.12"` and write `version: 1.0.0` as a plain semver (`1.0` is a YAML float and is refused).
@@ -517,6 +521,16 @@ Catalog of this release: {{tool_counts}}.
   an error, so check `status`. Each call costs about $0.0002 from the run's budget, and a run gets at most 200.
   Private, internal and metadata addresses, ports other than 80/443/8080/8443, and non-http(s) URLs answer
   `URL_REFUSED`. Prefer it over a `proxy_url` secret input: the person supplies nothing.
+- `browser.read` and `browser.act` work as the person on a site they signed in to once on Gravity (My agents,
+  Accounts). Declare the site under `accounts` (`linkedin.com` or `x.com`); the person's login and session stay in
+  Gravity's vault and browser worker, and agent code never sees them. `browser.read {account, url}` returns `url`,
+  `text` and `links` for a page of that account's own site only. `browser.act {account, action, target, text}` does one
+  typed action: `send_dm` (a profile URL or handle; LinkedIn only to first-degree connections), `connect` (a LinkedIn
+  profile URL, note at most 300 characters), `comment` (a LinkedIn post URL), `reply` (an X post URL). It must be under
+  `approvals.required_for`, so the person sees and approves each message; the gateway paces it and holds daily caps per
+  account (LinkedIn: 25 DMs, 15 notes, 10 comments; X: 15 DMs, 10 replies). Refusals come back as text codes:
+  `NOT_SIGNED_IN` (ask the person to sign in again, under "could not do"), `NOT_CONNECTED`, `DAILY_CAP`,
+  `BLOCKED_BY_SITE`. Browser time is charged to the run's budget.
 - Outbound HTTPS is open, so code may read public pages and APIs directly (`httpx`, `curl_cffi`). Anything that acts
   on an account still goes through a tool, so the person's connection and approvals apply.
 - Locally, `gravity connect <app>` attaches **your** account to your local runs (`gravity connect` alone lists what
@@ -818,6 +832,7 @@ checkpointing, durable approvals, `gravity.ask()`, action buttons, chains and ou
 | `APPROVAL_REQUIRED` (inside a hosted run) | A gated tool was called through the plain MCP client | Load tools with `await gravity.tools()` (section 5) |
 | `INVALID_APPROVAL`, `INVALID_TRIGGER`, `INVALID_KIND`, `INVALID_OUTPUT` | `required_for` names an undeclared tool; nothing can start the agent, a bad cron, both day fields, faster than 5 minutes; a monitor without a trigger; a bad output name or destination | Section 4 |
 | `INVALID_CONDITION`, `INVALID_ACTION`, `INVALID_MEMORY` | a `show_if` on a later input or a value it cannot take, a `choose_one` member required alone; an action that requires or chains to an undeclared one or loops, no default action; `memory` beside `state`, a bad collection | Section 4 |
+| `INVALID_ACCOUNT` | an account on a site other than linkedin.com or x.com, a name used twice or not an identifier; `browser.*` declared with no account to act through | Section 6 |
 | `UNDECLARED_TOOL_CALL` | Source names a catalog write tool the manifest does not grant | Declare and gate it, remove the call, or make a deliver-only tool an output `destination` |
 | `FILE_INPUT_UNSUPPORTED` (hosted build only) | A required `file` input | Make it optional, add a text fallback |
 | `SECRET_FOUND` (hosted build only) | The secret scanner matched something in the upload | Remove it, fixtures included; rotate it if it was real |

{gravity_cli-0.7.3 → gravity_cli-0.7.5}/uv.lock

@@ -164,7 +164,7 @@ wheels = [
 
 [[package]]
 name = "gravity-cli"
-version = "0.7.3"
+version = "0.7.5"
 source = { editable = "." }
 dependencies = [
     { name = "gravity-schema" },
@@ -201,7 +201,7 @@ dev = [
 
 [[package]]
 name = "gravity-schema"
-version = "0.7.3"
+version = "0.7.5"
 source = { editable = "packages/gravity-schema" }
 dependencies = [
     { name = "pydantic" },