Feature Request: Agent-scoped skills + agent-bound cron jobs

[srxly888-creator]

Feature Request: Agent-Scoped Skills + Agent-Bound Cron Jobs

Background

Current behavior in ClawX/OpenClaw integration is effectively global for skills and implicit for cron agent targeting:

• Skills enable/disable uses skills.update with only skillKey + enabled.
• Cron jobs created from ClawX use payload.kind = agentTurn and sessionTarget = isolated, but no explicit agentId is configured in UI/API.

For one-agent setups this is acceptable, but for multi-agent setups (e.g. 7-10 skills split by intent/purpose/goal) this causes:

• skill pollution (all agents see same skill set),
• weak role boundaries,
• less predictable automation behavior.

Proposal

Introduce two-level skill scope and explicit cron agent binding:

1. Global Skills (shared baseline)

• Available to all agents by default.
• Good for universal utilities (e.g. docs parsing).

2. Agent Skills (agent override layer)

• Per-agent allow/deny list on top of global.
• Supports role-specific skill sets (research/coding/review/ops).

3. Agent-Bound Cron

• Each cron job explicitly stores agentId.
• Runtime executes job in that agent context.

Suggested Data Model

type SkillPolicy = {
  globalEnabled: string[];          // skill keys
  agentOverrides: Record<string, {  // agentId
    enabled?: string[];             // explicit add
    disabled?: string[];            // explicit remove
  }>;
};

type CronJobExtension = {
  agentId: string;                  // default "main"
};

Effective skills for an agent:

effective(agent) = (globalEnabled - disabled(agent)) U enabled(agent)

UI/UX Requirements

1. Skills page:

• Scope switch: Global / Agent.
• When Agent selected, choose target agent and edit overrides.
• Show "effective" badge for current selected agent.

2. Agents page:

• Optional summary card: enabled skill count + overridden skills.

3. Cron page:

• Add Run as Agent selector in create/edit dialog.
• List view shows bound agent badge.

API Requirements (Host API / Main process)

1. Skills

• GET /api/skills/policy
• PUT /api/skills/policy/global
• PUT /api/skills/policy/agents/:agentId
• Keep existing skills.update for backward compatibility.

2. Cron

• Extend create/update payload with optional agentId.
• Persist and return agentId in /api/cron/jobs.

Backward Compatibility

• Existing installations default to:
  • all currently enabled skills -> globalEnabled
  • no per-agent overrides.
• Existing cron jobs default agentId = "main".
• If backend/runtime doesn’t support agent-bound execution yet:
  • preserve field in ClawX metadata and surface warning,
  • do not silently drop agent binding in UI.

Acceptance Criteria

1. A skill can be globally enabled but disabled for a specific agent.
2. A skill can be globally disabled but enabled for a specific agent.
3. Cron job can be created/edited with agentId.
4. Cron run history/session key resolves to that agent context consistently.
5. Existing users upgrade without behavior regression.

Why this matters

• Better separation of intents/purposes/goals per agent.
• Lower accidental tool invocation risk.
• Better reproducibility for scheduled workflows.
• Foundation for policy controls (team/workspace governance).

Implementation Notes

Likely needs cross-repo coordination:

• ClawX UI + Host API changes (this repo).
• OpenClaw gateway/runtime support for agent-scoped skill resolution and cron agent execution.

If runtime support is not available, ship in phases:

1. ClawX-side model + UI + persistence + warnings.
2. Runtime wiring once OpenClaw supports agent-scoped resolution.

[srxly888-creator]

Upstream runtime tracking issue has been created:

• Runtime contract for cron agentId + delivery(channel/to) semantics openclaw/openclaw#48750

Why this matters for #543:

• ClawX can already send agentId and delivery fields,
• but deterministic behavior depends on OpenClaw gateway/runtime semantic guarantees.

Suggested collaboration split:

1. Keep ClawX UI/API implementation moving here (field modeling, warnings, UX).
2. Track runtime semantic support and conformance in upstream issue #48750.
3. Link PRs from both sides to avoid drift.

[srxly888-creator]

Upstream runtime tracking has now been split into two issues for parallel delivery:

1. Cron execution + delivery semantics (agentId, delivery, channel, optional to):

• Runtime contract for cron agentId + delivery(channel/to) semantics openclaw/openclaw#48750

2. Global skills + agent override + effective skill resolution:

• Runtime support for global skills + agent override (effective skill resolution) openclaw/openclaw#48752

Proposed cross-repo split:

• ClawX: UI/API/persistence/warnings and backward-compatible UX.
• OpenClaw: runtime semantic guarantees + conformance tests.

This should reduce ambiguity and let teams work in parallel without contract drift.

[srxly888-creator]

Implemented ClawX-side skill policy scope support and opened PR: https://github.com/ValueCell-ai/ClawX/pull/550\n\nThis covers Global + Agent override policy model/routes/UI (with effective badges and runtime warning), plus i18n/docs/tests updates.

[srxly888-creator]

Checked #543 follow-up status:

• Active implementation PR: feat(skills): add global/agent policy scope for issue #543 #550
• Re-validation completed (typecheck + targeted tests + full unit suite) and posted on PR thread.

No additional reviewer comments are currently pending on #543/PR #550 at this time. Will continue addressing immediately once maintainers leave concrete review items.

[vcfgv]

Thanks for the solid work on this issue and for documenting it clearly. Appreciate the contribution.