Bolta is a policy-aware creative workflow runtime.

This Skills Pack transforms Bolta from a scheduling tool into an agent-native content operating system — where agents generate, route, review, and automate content safely inside governed workspaces.

Agents act.
Bolta enforces.
Humans retain control.

V2 introduces agentic job execution:

• Jobs replace RecurringTemplates (agents reason about tasks instead of filling templates)
• Agent memory and learning (agents improve over time)
• Agent hiring flow (marketplace-driven onboarding)
• @Mention review workflow (feedback loops)
• Enhanced business context (products, audience DNA, performance data)

Core philosophy: Skills are capabilities that agents choose based on reasoning, not pipeline steps.

Different agents use the same skills differently based on their role and context.

V2 introduces specialized agent types with different mindsets:

Bolta supports five foundational workflows.

Purpose
Agents generate content using jobs (not templates), then route to Inbox Review.

What Happens

• Agent receives job brief (natural language instructions)
• Agent reasons about task and chooses tools
• Agent drafts content in Brand Voice
• Content automatically routes to Inbox for human review
• Routing follows workspace policy automatically

Policy Enforcement

• Safe Mode ON → always routes to Inbox Review
• Safe Mode OFF → defaults to Inbox unless role is Editor/Admin
• Role capabilities enforced server-side

Skills Used

• bolta.draft.post
• bolta.get_voice_profile
• bolta.list_recent_posts
• bolta.agent.memory (remember/recall)

V2 Enhancement: Agents learn from feedback and adapt drafts over time.

Purpose
Run scheduled content generation automatically.

What Happens

• Job fires on schedule (cron, event trigger, or manual)
• Agent loads context (voice, memory, business DNA)
• Agent generates content
• Either routes to review or schedules directly (based on policy)

Hard Constraints

• Safe Mode must be respected
• Scheduling requires explicit permission
• Role capabilities enforced server-side

Skills Used

• bolta.job.execute
• bolta.draft.post
• bolta.get_voice_profile

V2 Enhancement: Jobs use agent reasoning instead of template fills.

Purpose
Human or agent reviews pending content before approval/scheduling.

What Happens

• Drafts accumulate in Inbox
• Reviewer (human or agent) triages items
• Items approved/rejected with feedback
• Approved items scheduled or published

Skills Used

• bolta.inbox.triage
• bolta.review.approve_and_route
• bolta.reject_post
• bolta.add_comment

V2 Enhancement: Reviewer agents can auto-triage with confidence scores.

Purpose
Onboard agents as teammates from marketplace presets.

What Happens

• User selects agent preset (e.g., "The Hype Man")
• Agent configuration loads (persona, model tier, skills)
• Agent generates preview draft
• User reviews preview and activates agent
• Agent jobs become active

Skills Used

• bolta.agent.hire
• bolta.agent.configure
• bolta.agent.activate_job

Purpose
Capture feedback on live posts to improve future content.

What Happens

• User mentions agent on published post (@AgentName)
• Mention captured as feedback signal
• Agent stores learning in memory
• Future drafts incorporate feedback

Skills Used

• bolta.agent.mention (read-only)
• bolta.agent.memory (remember)

Bolta is divided into skill planes. Skills orchestrate workflows; the backend enforces identity, permissions, and routing.

A first-class actor in the system.

• type: human | agent
• Owns API keys / auth tokens
• Produces auditable actions

The governance boundary.

• Policies (Safe Mode, Inbox Direct Scheduling, etc.)
• Memberships (principals + roles)
• Resources (voices, templates, posts, jobs, accounts)

Roles determine what a principal can do.

• Viewer / Creator / Editor / Admin
• Capabilities map to actions (create, submit, approve, schedule, publish)

A centralized decision layer used by every entrypoint (UI, API, MCP tools):

• authorize(principal, workspace, action, resource) -> allow/deny (+reason)
• resolve_route(policy, action, requested_state) -> final_state

draft
  → inbox (submit_for_review OR Safe Mode routing)
  → approved (Editor/Admin approval)
  → scheduled (if allowed)
  → published

Failure routing:

• schedule/publish denied or fails → inbox with warning (preferred)
• hard failures → draft + error metadata

V2 Note: State names updated (pending_review → inbox in V2 codebase, but conceptually identical).

Voice is a first-class primitive in Bolta.

Before content can be generated, a workspace must have an active Voice Profile. Voice Profiles define tone, structure, vocabulary bias, and stylistic rules that agents must follow.

Rule: Before executing any Content Plane skill, if no active voice_profile, require bolta.voice.bootstrap.

Voice is a hard dependency for content generation.

Skills:

• bolta.voice.bootstrap — Create comprehensive voice profile
• bolta.voice.learn_from_samples — Refine voice from existing content
• bolta.voice.evolve — Evolve voice based on performance data and feedback
• bolta.voice.validate — Validate content against voice profile rules

Creates and shapes content.

Responsibilities:

• Voice application (tone, style rules)
• Template rendering (V1 compatibility)
• Draft generation
• Job execution (V2)

Primary primitives:

• Voice Profiles
• Templates (deprecated in V2, use jobs)
• Draft Posts
• Jobs (V2)

Typical states:

• draft
• inbox (V2)

Skills:

• bolta.draft.post — Create draft post in brand voice
• bolta.get_voice_profile — Load voice details
• bolta.list_recent_posts — Check recent content
• bolta.get_business_context — Load products, audience, brand DNA (V2)
• bolta.week.plan — Content calendar planning (V2)

Controls human oversight and safe routing.

Responsibilities:

• Inbox Review queue
• Bundling posts for review
• Approvals and edits
• Safe Mode enforcement

Primary primitives:

• Inbox Items
• Review Bundles
• Review Status transitions

Typical states:

• inbox
• approved
• needs_edits

Rule: If Safe Mode is ON, content must pass through this plane before schedule/publish.

Skills:

• bolta.inbox.triage — Inbox management and prioritization (V2)
• bolta.review.approve_and_route — Approve and schedule
• bolta.reject_post — Reject with feedback
• bolta.add_comment — Add review notes

Runs recurring workflows and conditional scheduling.

Responsibilities:

• Job-triggered generation (V2)
• Cron-triggered generation (V1 compatibility)
• Scheduling posts (only if allowed)
• Retry logic + idempotency keys
• Backoff + failure routing to review

Primary primitives:

• Jobs (V2)
• Job Runs (V2)
• Generation Tasks
• Scheduling Requests

Typical states:

• scheduled
• failed → inbox

Skills:

• bolta.job.execute — V2 job execution engine (documentation)
• bolta.cron.generate_and_schedule — Cron-based automation (V1 compatibility)
• bolta.loop.from_template — DEPRECATED (V1 legacy, see DEPRECATED.md)

Agent lifecycle, orchestration, and memory.

Responsibilities:

• Agent hiring and configuration
• Agent memory (cross-run learning)
• @Mention feedback loops
• Job activation

Primary primitives:

• Agent Teammates
• Agent Memory
• Agent Jobs
• Mentions

Skills:

• bolta.agent.hire — Conversational agent onboarding
• bolta.agent.configure — Modify agent settings
• bolta.agent.activate_job — Activate paused jobs
• bolta.agent.memory — remember/recall for learning
• bolta.agent.mention — @mention handling (read-only)

Performance tracking, audience insights, data-driven recommendations.

Skills:

• bolta.get_post_metrics — Post performance data
• bolta.get_audience_insights — Demographics, behavior patterns
• bolta.get_best_posting_times — Optimal schedule based on YOUR data

Community management, mentions, replies, moderation.

Skills:

• bolta.get_mentions — Fetch mentions and replies
• bolta.reply_to_mention — Draft replies in brand voice
• bolta.get_comments — Fetch comments for moderation

Governance, identity, security, and observability.

Responsibilities:

• Principals (humans + agents)
• Workspace memberships
• Role assignment
• API keys (scoped)
• Audit logs
• Policy configuration

Primary primitives:

• Principals
• Memberships
• Policies
• Keys
• Audit Events

Skills:

• bolta.workspace.config — Workspace settings
• bolta.policy.explain — Explain workspace policies
• bolta.audit.export_activity — Export activity logs
• bolta.team.create_agent_teammate — Create agent service account (V2)
• bolta.team.rotate_key — Rotate agent API keys
• bolta.quota.status — Check workspace quota and rate limits

All skills rely on a consistent, policy-aware tool layer:

Policy & Capabilities:

• bolta.get_workspace_policy
• bolta.get_my_capabilities

Voice & Context:

• bolta.get_voice_profile
• bolta.get_business_context (V2)
• bolta.list_templates
• bolta.render_template

Content Operations:

• bolta.draft_post (V2, replaces create_post)
• bolta.create_loop (V1 compatibility)
• bolta.list_recent_posts (V2)

Review Operations:

• bolta.inbox.triage (V2)
• bolta.approve_post
• bolta.reject_post
• bolta.add_comment (V2)
• bolta.schedule_post
• bolta.publish_post

Agent Operations (V2):

• bolta.agent.hire
• bolta.agent.configure
• bolta.agent.activate_job
• bolta.agent.memory (remember/recall)

Control Plane:

• bolta.team.create_agent_teammate
• bolta.workspace.config
• bolta.policy.explain
• bolta.audit.export_activity

Every skill must:

1. Query workspace policy
2. Query effective capabilities
3. Respect Safe Mode
4. Never bypass server authorization

Bolta scales from assisted generation to fully autonomous runtime. These levels describe how customers typically adopt Bolta over time.

• Draft generation
• Job creation (V2)
• Weekly planning
• Human approval required

• Inbox triage
• Review digest
• Controlled approval routing
• Optional conditional scheduling

• Job-based generation (V2)
• Conditional scheduling
• Policy-aware automation
• Agent learning from feedback

• Agent management
• Key rotation
• Workspace policy controls
• Audit exports
• Activity filtering

Bolta is not:

• An API wrapper
• A content generator
• A basic scheduler

Bolta is: A policy-aware agent runtime for creative systems.

Agents generate.
Policies govern.
Humans approve.

V2 Addition: Skills are capabilities agents choose based on reasoning, not pipeline steps. The same skill serves different purposes depending on the agent's role and context.

This is how autonomous content becomes safe, scalable, and enterprise-ready.

Each skill includes:

Required sections:

• YAML frontmatter (name, version, description, category, roles_allowed, inputs_schema, outputs_schema)
• Goal — What the skill does (1-2 sentences)
• Steps — Numbered workflow
• Hard rules — Must/must not constraints
• Failure handling — What happens when things go wrong
• Output — Return format

V2 additions (optional):

• Which agent types use this — Brief context
• Example usage — Real-world scenarios

Every skill has a schema.json file (100% coverage across all 38 skills).

• name — Matches server-side tool definition
• description — How agents understand the tool
• input_schema — Parameters with descriptions
• required — Mandatory fields

Shows how the tool connects to data (not always included).

If you're using RecurringTemplates (V1 loops):

1. Read: skills/automation-plane/bolta.loop.from_template/DEPRECATED.md
2. For each template:
   • Create Content Creator agent via bolta.agent.hire
   • Create Job with natural language instructions (not template variables)
   • Bind voice_profile_id, account_ids
   • Activate job via bolta.agent.activate_job
3. Optional: Delete old template
4. Observe improvement: V2 agents adapt and improve over time

V1 skills still work for backward compatibility, but V2 jobs are recommended.

1. Choose the right plane (content, review, agent, analytics, etc.)
2. Create directory: skills/{plane}/bolta.skill_name/
3. Write SKILL.md:
   • Start with YAML frontmatter
   • Write concrete steps (numbered)
   • Add hard rules and failure handling
   • Optionally add V2 agent context (brief)
4. Write schema.json — Match server-side tool definition
5. Implement handler — Add to server tool registry
6. Update this README — Add to plane documentation

Use the Bolta API or MCP server to test skills in context:

# Create test agent
curl -X POST /api/v1/agents \
  -d '{"type": "content_creator", "name": "Test Agent"}'

# Create test job
curl -X POST /api/v1/jobs \
  -d '{"agent_id": "...", "run_instructions": "Test task"}'

# Trigger execution
curl -X POST /api/v1/jobs/{job_id}/run

1. Create workspace and connect social accounts
2. Set up voice profile via bolta.voice.bootstrap
3. Hire an agent from the marketplace (optional, V2 feature)
4. Create drafts via bolta.draft.post or job execution
5. Review in Inbox and approve/reject with feedback
6. Schedule or publish approved content
7. Watch agents improve as they learn from feedback (V2)

# Draft a post
bolta.draft.post({
  workspace_id: "...",
  voice_profile_id: "...",
  prompt: "Write about our new feature"
})

# Review and approve
bolta.review.approve_and_route({
  workspace_id: "...",
  post_ids: ["..."],
  schedule_mode: "use_suggested_time"
})

This is what policy-aware agentic automation looks like.

Agents act. Bolta enforces. Humans retain control.

Welcome to Bolta Skills Pack V2. 🚀