roboto.domain.skills
Stored, versioned procedures the chat AI can apply during a conversation.
A skill is text instructions the model follows when invoked — either manually via a chat-composer chip or automatically when the model selects it from the load_skill tool’s registry. Skills are owned by a user, optionally shared with the user’s organization, and versioned in place; each version is mutable without lifecycle states. Visibility and edit rights follow the skill’s accessibility: private (author-only), org (visible org-wide, author-only to edit), or org-editable (visible org-wide and editable by any subscribed member). Per-user subscriptions carry an ai_version pin that controls whether — and which version — the AI auto-invoke registry exposes to the subscriber.
Submodules
Package Contents
CreateSkillRequest
Bases: pydantic.BaseModel
Create a new skill plus its first version (version=1) in one call.
The author is auto-subscribed at creation time with ai_version set to 1 so the skill is immediately offered to the author’s AI.
Parameters
data AnyAttributes
CreateSkillRequest.accessibility
Visibility scope: Private (only the author), Org (visible to every org member, author-only to edit), or OrgEditable (visible to every org member, editable by any member who subscribes). The author can flip this later via UpdateSkillMetadataRequest.
CreateSkillRequest.body
Procedure text v1 — the instructions the model executes when this version is invoked. Bumping the version creates a new row; this field never re-edits v1’s body after the fact (use UpdateSkillVersionRequest for that).
CreateSkillRequest.description
“When to use” text for v1. Surfaces verbatim in the load_skill tool’s description so the model can decide whether to invoke this skill. Bumping the version replaces this text on the new row; this field never re-edits v1’s description after the fact.
CreateSkillRequest.model_config
model_config #Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].
CreateSkillRequest.name
Skill name. Unique per org across every org-visible skill (Org and OrgEditable share one name namespace); private skills are unique per (org_id, created_by). A user’s private skill and an org-shared skill may therefore share a name without conflict. Must match SKILL_NAME_PATTERN so the name is parseable inside the chat composer’s /slug token (no whitespace, letters/digits/hyphens/underscores only).
CreateSkillRequest.relevant_topics
Topic names v1’s procedure investigates (e.g. "/imu/data"), supplied by the caller. Version-scoped content — see SkillVersionRecord.relevant_topics. The SDK stores the list as given; it performs no extraction. (The Roboto web UI populates it by scanning the body for [[topic]] references, but that is a UI convenience, not an SDK guarantee — a direct caller passes whichever names the procedure investigates.) Edit later via UpdateSkillVersionRequest.relevant_topics.
CreateSkillRequest.tags
Initial set of tags. Edits after creation flow through UpdateSkillMetadataRequest.put_tags / .remove_tags so concurrent updates merge cleanly — see SkillRecord.tags.
CreateSkillVersionRequest
Bases: pydantic.BaseModel
Add a new version to an existing skill. The new version is assigned MAX(version) + 1.
Parameters
data AnyAttributes
CreateSkillVersionRequest.body
Procedure text the model executes when this version is invoked.
CreateSkillVersionRequest.description
“When to use” text for this version. Surfaces verbatim in the load_skill tool’s description on every turn this version is offered to the AI.
CreateSkillVersionRequest.model_config
model_config #Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].
CreateSkillVersionRequest.relevant_topics
Topic names this version’s procedure investigates. Version-scoped content — see SkillVersionRecord.relevant_topics.
MAX_SKILL_DESCRIPTION_LENGTH
SKILL_NAME_PATTERN
SetSkillSubscriptionRequest
Bases: pydantic.BaseModel
Upsert the caller’s subscription for a skill.
Idempotent: creates the subscription row if missing, otherwise updates ai_version in place. Visibility-gated only — the caller does not need to be the skill’s author. The server enforces that ai_version, if provided, references an existing version of the skill.
Parameters
data AnyAttributes
SetSkillSubscriptionRequest.ai_version
Pinned AI-available version. None means “subscribed, but not exposed to AI”.
SetSkillSubscriptionRequest.model_config
model_config #Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].
Skill
An AI skill — a versioned, accessibility-scoped procedure the chat AI can apply.
Use create() to create a new skill, from_id() / from_name() to load existing ones, and list_for_org() to iterate. The constructor is internal.
Parameters
roboto_client roboto.Properties
Skill.accessibility
Skill.create()
Create a new skill plus its first version (version=1).
The skill is created in the caller’s organization. With the default SkillAccessibility.Private accessibility only the caller can see, edit, or delete it. SkillAccessibility.Org makes it readable by all org members (author-only to edit); SkillAccessibility.OrgEditable additionally lets any member who subscribes edit its versions, name, and tags. The author is auto-subscribed at creation time and the new version is pinned as Available to AI. Other org members must subscribe() to see the skill in their AI auto-invoke registry. Pass tags to seed the skill’s tag list at creation time; later edits flow through UpdateSkillMetadataRequest. Pass relevant_topics to record the topic names v1’s procedure investigates; the chat AI resolves them to schemas when it loads the skill (see SkillVersionRecord.relevant_topics).
Usage
skill = Skill.create(
name="qa-review",
description="Run when the user asks for a QA review of a dataset.",
body="Step 1: load the dataset summary...",
accessibility=SkillAccessibility.Org,
tags=["qa-review", "triage"],
relevant_topics=["/imu/data", "/gps/fix"],
)
skill.skill_id
# 'sk_...'Parameters
name strdescription strbody straccessibility roboto.tags Optional[collections.relevant_topics Optional[collections.caller_org_id Optional[str]roboto_client Optional[roboto.Return type
Skill.create_version()
Add a new version to this skill. The server assigns MAX(version) + 1.
Permitted for the author and — on an SkillAccessibility.OrgEditable skill — for any subscribed org member. Subscribers who pinned the previous version stay on that pin until they explicitly re-pin — the new row does not auto-promote.
Pass relevant_topics to record the topic names this version’s procedure investigates; the chat AI resolves them to schemas when it loads the skill (see SkillVersionRecord.relevant_topics).
Usage
v2 = skill.create_version(
CreateSkillVersionRequest(
description="Run when ...",
body="Updated procedure ...",
relevant_topics=["/imu/data", "/gps/fix"],
)
)
v2.version
# 2Parameters
Properties
Skill.delete()
Hard-delete this skill. Author-only, including on OrgEditable skills.
Cascades to all versions and subscriptions. Existing chats keep any fabricated load_skill tool_use / tool_result blocks they’ve already produced — the body is captured at invocation time and written into the transcript.
Usage
skill.delete()Return type
Skill.delete_version()
Delete a single version. If it’s the last remaining version, the parent skill is removed too.
Permitted for the author and — on an SkillAccessibility.OrgEditable skill — for any subscribed org member. A subscribed non-author may not delete the last remaining version: that cascades into a full skill delete, which is author-only. The author has no such restriction.
Subscriptions pinned to the deleted version have their ai_version nulled out via a server-side trigger; the subscription row survives.
Usage
skill.delete_version(1)Parameters
version intReturn type
Skill.from_id()
Load a skill by its skill_id.
Parameters
skill_id strroboto_client Optional[roboto.Raises
No skill with this id exists, or the caller cannot see it (visibility-gated).
Return type
Usage
skill = Skill.from_id("sk_abc123")
skill.name
# 'qa-review'Skill.from_name()
Load a skill by name in the caller’s organization.
Skill names are unique within (org_id, accessibility). When both a private and an org-shared skill share a name in the same org, this method returns the caller’s private one (the manual-invocation tie-break — see SkillsRepo.get_skill_by_name()).
Parameters
name strSkill name. URL-encoded by the SDK.
caller_org_id Optional[str]Look up in this org. Defaults to the caller’s current org.
roboto_client Optional[roboto.Raises
No skill with this name is visible to the caller in the target org.
Return type
Usage
skill = Skill.from_name("qa-review")
skill.accessibility
# <SkillAccessibility.Org: 'org'>Skill.get_summary()
Load one skill’s summary: the skill, its latest version, and your subscription.
The single-skill counterpart of list_for_org(). Prefer it over filtering that listing when you already know the skill_id — it costs one row instead of the whole org’s skills.
Parameters
skill_id strroboto_client Optional[roboto.Raises
No skill with this id exists, or the caller cannot see it (visibility-gated, same as from_id()).
Return type
Usage
summary = Skill.get_summary("sk_abc123")
summary.latest_version.version
# 3
summary.subscription is None # not subscribed
# TrueSkill.get_version()
Load a specific version of this skill.
Parameters
version intRaises
No such version exists.
Usage
v2 = skill.get_version(2)
print(v2.body)Skill.list_for_org()
Yield SkillSummary items for every skill the caller can see in their org.
The summary includes the latest version (MAX(version)) and the caller’s own subscription row when one exists. When scope is provided the result is restricted to either Personal (authored or subscribed) or Org (org-shared skills the caller did not author, regardless of subscription state). Omit scope to receive every visible skill in one stream.
Usage
List the caller’s Personal-tab skills:
for summary in Skill.list_for_org(scope=SkillListScope.Personal):
print(summary.skill.name, summary.subscription.ai_version if summary.subscription else None)Iterate every visible skill (Personal + Org), in one pass:
all_visible = list(Skill.list_for_org())Parameters
scope Optional[roboto.caller_org_id Optional[str]roboto_client Optional[roboto.Return type
Skill.list_known_tags()
Return the distinct tags found on skills the caller can see in this org.
Visibility-filtered the same way list_for_org() is — private skills owned by other users contribute no tags. Suitable for powering a tag-autocomplete UI.
Usage
Skill.list_known_tags()
# ['qa-review', 'triage', 'experiments']Parameters
caller_org_id Optional[str]roboto_client Optional[roboto.Return type
Skill.list_versions()
List every version of this skill, newest first.
Usage
for version in skill.list_versions():
print(version.version, version.description)Return type
Properties
Skill.record
Skill.refresh()
Re-fetch this skill’s record from the server and return self.
Useful after another caller may have updated the skill — bumps the local view past stale data.
Usage
skill.refresh()
skill.name # now reflects any server-side rename
# 'qa-review'Return type
Skill.set_ai_version()
Pin (or clear) which version of this skill the AI auto-invokes for the caller.
Pass an integer to expose that exact version to AI auto-invocation; pass None to disable AI auto-invocation while keeping the subscription. Implicitly subscribes the caller if no row exists yet. Visibility-gated; the caller does not have to be the author.
Usage
Pin version 2 for AI auto-invoke:
skill.set_ai_version(2)Stop the AI from auto-invoking this skill, but stay subscribed so manual chip-invocation still works:
skill.set_ai_version(None)Parameters
version Optional[int]Properties
Skill.subscribe()
Subscribe to this skill — adds it to the caller’s Personal tab.
Idempotent: if the caller is already subscribed (or is the author), the existing row is returned unchanged. New subscriptions default to ai_version=None (not yet exposed to AI). Use set_ai_version() to enable AI auto-invocation.
Usage
sub = skill.subscribe()
sub.ai_version is None
# TrueProperties
Skill.unsubscribe()
Remove the caller’s subscription row — removes the skill from their Personal tab.
No-op if there is no subscription. Authors who unsubscribe from their own skill can re-subscribe later; authorship is unchanged.
Usage
skill.unsubscribe()Return type
Skill.update_metadata()
Update skill-level metadata (name, accessibility, tags).
Editing the name and tags is permitted for the author and — on an SkillAccessibility.OrgEditable skill — for any subscribed org member. Changing accessibility is always author-only.
Updates apply in place; the local record is replaced with the server’s authoritative copy.
Usage
skill.update_metadata(
UpdateSkillMetadataRequest(
accessibility=SkillAccessibility.Org,
put_tags=["qa-review"],
)
)Parameters
Return type
Skill.update_version()
Edit fields on an existing version in place.
Permitted for the author and — on an SkillAccessibility.OrgEditable skill — for any subscribed org member.
Subscribers pinned to this version see the new body on their next AI invocation; there is no per-edit revision. Use create_version() when callers should not be auto-migrated.
Usage
skill.update_version(2, UpdateSkillVersionRequest(body="..."))Parameters
version intSkillAccessibility
Bases: roboto.compat.StrEnum
Controls who can see and edit a skill.
Attributes
SkillAccessibility.Org
All members of the owning org can see and invoke the skill. Only the author can edit.
SkillAccessibility.OrgEditable
All members of the owning org can see and invoke the skill, and any member who has subscribed to it can also edit its versions, name, and tags. Changing the skill’s accessibility and deleting the whole skill remain author-only.
SkillAccessibility.Private
Only the author (created_by) can see, edit, or invoke the skill.
SkillListScope
Bases: roboto.compat.StrEnum
Which caller-relative slice of the visible skills a list-skills query targets.
SkillRecord
Bases: pydantic.BaseModel
Top-level skill identity, ownership, and visibility.
A skill on its own carries no procedure content — see SkillVersionRecord.
Parameters
data AnyAttributes
SkillRecord.accessibility
SkillRecord.created
SkillRecord.created_by
SkillRecord.modified
Last edit timestamp. Skill mutations are author-only, so the editor is always created_by — no separate modified_by field is stored.
SkillRecord.name
SkillRecord.org_id
SkillRecord.skill_id
SkillRecord.tags
Free-form labels for categorization (e.g. "qa-review", "experiments"). Skill-level metadata, not version-scoped — they describe what the skill is for, not how a specific version implements it. Edited via the changeset fields on UpdateSkillMetadataRequest (put_tags / remove_tags).
SkillSubscriptionRecord
Bases: pydantic.BaseModel
Per-user state for a single skill.
Created when a user subscribes to a skill (or authors one — authors are auto-subscribed at create time). Carries the user’s choice of which version, if any, should be exposed to the AI auto-invoke registry. ai_version=None means the user is subscribed but has not enabled the skill for AI auto-invoke; manual chip-invocation still works.
Parameters
data AnyAttributes
SkillSubscriptionRecord.ai_version
If set, the AI’s LoadSkillTool registry surfaces this exact version of the skill to this user. If None, the skill is hidden from AI auto-invocation for this user. Manual invocation works regardless.
SkillSubscriptionRecord.skill_id
SkillSubscriptionRecord.user_id
SkillSummary
Bases: pydantic.BaseModel
Compact projection of a skill plus its latest version (if any).
Also includes the caller’s personal subscription state for the skill, when one exists. subscription=None means the caller has neither authored nor subscribed to the skill.
Parameters
data AnyAttributes
SkillSummary.latest_version
The MAX(version) row for this skill. Always populated for summaries returned from the public API: Skill.create() makes the skill row and v1 atomically, and deleting the last version cascades to delete the parent skill in the same transaction. The Optional is defense-in- depth for repo-direct callers (test fixtures, manual SQL) and should not be treated as a real branch by SDK consumers.
SkillSummary.skill
SkillSummary.subscription
The caller’s subscription row for this skill, if any. Populated by the server based on the caller’s identity; not used in INSERT or UPDATE payloads.
SkillVersionRecord
Bases: pydantic.BaseModel
A single, mutable version of a skill’s content.
Parameters
data AnyAttributes
SkillVersionRecord.created
SkillVersionRecord.description
Short ``when to use’’ text. Surfaces verbatim in the LoadSkillTool description.
SkillVersionRecord.modified
SkillVersionRecord.relevant_topics
Topic names this version’s procedure investigates (e.g. "/imu/data").
Version-scoped content, not skill-level metadata — they track the body, so a new version may reference a different set. When the chat AI loads this skill and is given a dataset scope, it resolves each name to a topic schema and returns them alongside the body in the same load_skill tool result, sparing a follow-up get_topic_schema turn. Names only; field-level references live in the body text. Empty when the author listed none.
SkillVersionRecord.skill_id
SkillVersionRecord.version
UpdateSkillMetadataRequest
Bases: pydantic.BaseModel
Update top-level skill identity.
Editing name / put_tags / remove_tags is permitted for the author and — on an OrgEditable skill — for any subscribed org member. Changing accessibility is always author-only.
Parameters
data AnyAttributes
UpdateSkillMetadataRequest.accessibility
New visibility scope. Omit to leave unchanged. Author-only — a subscribed non-author editor of an OrgEditable skill cannot change it. Flipping Private → Org / OrgEditable immediately exposes the skill to every org member; flipping to Private prunes every non-author subscription.
UpdateSkillMetadataRequest.model_config
model_config #Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].
UpdateSkillMetadataRequest.name
New skill name. Omit (the NotSet default) to leave unchanged. Must match SKILL_NAME_PATTERN and stays unique per org (private skills per author).
UpdateSkillMetadataRequest.put_tags
Tags to add. Empty (the default) means “no change to tags.” Additive semantics — does not clear other tags.
UpdateSkillMetadataRequest.remove_tags
Tags to remove. Empty (the default) means “no change to tags.” Removing a tag the skill doesn’t have is a no-op.
UpdateSkillVersionRequest
Bases: pydantic.BaseModel
Edit fields on an existing version (mutates in place).
Parameters
data AnyAttributes
UpdateSkillVersionRequest.body
New procedure text. Omit (the NotSet default) to leave unchanged. The change applies in place to this version — subscribers pinned to it will see the new body on their next AI invocation; there is no per-edit revision.
UpdateSkillVersionRequest.description
New “when to use” text. Omit (the NotSet default) to leave unchanged.
UpdateSkillVersionRequest.model_config
model_config #Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].
UpdateSkillVersionRequest.relevant_topics
New topic-name list. Omit (the NotSet default) to leave unchanged; pass an empty list to clear all topics. Version-scoped content — see SkillVersionRecord.relevant_topics.