Agent Budget Governance
PassAudited by VirusTotal on May 11, 2026.
Overview
Type: OpenClaw Skill Name: agent-budget-governance Version: 1.0.0 The skill bundle implements a legitimate budget governance system for multi-agent environments, tracking token usage via local JSON files. The core logic in `scripts/budget_audit.py` is transparent, performing file-based audits and resets without any network access, shell execution, or unauthorized data access. The instructions provided in `SKILL.md` and `references/GOVERNANCE.md` are strictly aligned with the stated purpose of cost management and do not exhibit signs of malicious prompt injection or deceptive behavior.
Findings (0)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
You may believe runaway agent spending is automatically blocked when the supplied implementation mainly records status changes that other agents must voluntarily honor.
This claims direct mesh privilege revocation, but the included script only updates BUDGET.json status fields and audit logs; no provided code modifies openclaw.json or blocks spawning. Users could over-trust the skill as an enforced spending control.
Demotion removes an agent's `subagents.allowAgents` list in `openclaw.json`, preventing them from spawning other agents.
Treat this as advisory governance unless you verify actual spawn-blocking integration, or add reviewed code that updates the relevant OpenClaw privilege configuration safely and reversibly.
Budget enforcement may silently fail or use defaults if users create files from the documented example.
The documented budget file example uses field names that do not match the script's expected fields such as daily_limit_output_tokens, used_output_tokens, consecutive_overbudget_days, and status. A user following this example could get ineffective audits or missed thresholds.
"daily_limit": 50000, "used_today": 23450, "consecutive_over_days": 0, "demoted": false
Update the documentation and examples to match the script schema, or update the script to support the documented schema with validation and clear errors.
Running the audit can change persistent budget state for multiple agents at once.
The audit walks all agent directories and rewrites each agent's BUDGET.json. This is aligned with the governance purpose, but one bad budget file or schema mismatch can affect team-wide status and resets.
AGENTS_DIR = os.path.expanduser("~/.openclaw/workspace/agents") ... agent_dirs = sorted(glob.glob(os.path.join(AGENTS_DIR, "*"))) ... with open(budget_path, "w") as f: json.dump(budget, f, indent=2)Back up existing budget files, test with --json or a staging workspace first, and add validation/error handling before scheduling it automatically.
If scheduled, the audit will continue resetting counters and changing agent statuses without a manual run each time.
The skill is designed to be run periodically by a heartbeat or cron-like mechanism. This is disclosed and purpose-aligned, but it means the script can keep making persistent state changes after setup.
Heartbeat runs budget_audit.py
Only schedule it after confirming the schema, paths, and enforcement behavior match your environment, and keep a clear rollback process.