Bind 'Done' to Unfakeable Gates: Night-Market Completion-Integrity

Where this fits

Your autonomous agent—say, an egregore orchestrator or a herald auto-continue loop—reports "done" after a few turns, but the actual work is incomplete. The loop stops early, the user gets a half-finished result, and you're left debugging why the agent thought it was finished. This skill provides an executable campaign that binds the loop's done signal to unfakeable gates, forcing the agent to prove completion through observable checks before it's allowed to stop.

Why agents benefit

• Prevents premature termination: Stop hooks block the agent from ending its turn until all gate conditions pass, so "done" is earned, not assumed.
• Enforces verifiable checks: Each phase ends with expected observations and branch instructions, giving the agent concrete evidence to produce rather than vague self-assessment.
• Integrates with existing hooks: Works directly with Claude Code's Stop hook mechanism, printing approve or block decisions based on actual state.
• Promotes completion_integrity to default-on: The campaign's goal is to flip this flag from off to on, making integrity a system-wide default rather than an opt-in.
• Provides decision-gated phases: You can run phases sequentially, each with clear gates and branch instructions, so the agent knows exactly what to do next.

Practical scenarios

Egregore orchestrator loop

Your egregore loop summons sub-agents and coordinates tasks. Without integrity gates, it might stop after the first sub-agent returns, leaving others pending. This skill adds a gate that requires all sub-agents to report completion with evidence before the loop can approve a stop.

Herald auto-continue Stop hook

Herald's auto-continue hook fires when the agent tries to end its turn. By applying this campaign, you replace the default approve with a block unless the agent can show that all defined completion criteria—like file checksums or test results—are satisfied.

External ralph-wiggum loop

An externally installed loop that keeps working without human turns can drift into endless loops or early exits. This skill's phased gates give you a way to instrument that loop with the same integrity checks, so it only stops when the work is genuinely complete. n## Add it to your agent workflow Install the skill directly from the Night Market repository:

npx skills add athola/claude-night-market --skill night-market-completion-integrity-campaign

nThen run the campaign phases in order. For example, after installing, invoke the campaign's first phase against your loop's configuration:

# Example: run phase 1 with a custom stop hook
claude-night-market run phase-1 --hook /path/to/stop-hook.js

Each phase ends with a gate that prints expected observations and branch instructions, guiding you to the next phase.

Compare before adopting

Before integrating, compare this skill against night-market-validation-and-qa—the former is for completion integrity, the latter for QA gates, so pick based on your loop's failure mode. Assess the skill's maintenance freshness (check the GitHub repo for recent commits) and its fit with your existing hook setup. Also look at alternatives like custom stop-hook scripts or other integrity tools, and verify that the phased approach matches your workflow's complexity.

Why it is worth tracking

With 335 stars and an active author (athola), this skill shows real community momentum. The campaign's explicit goal—promoting completion_integrity to default-on—signals that it's solving a persistent pain point in autonomous agent development. Track it when you're building or debugging long-running loops, especially if you've seen premature stops or unverifiable completions. For a detailed breakdown, see its page on OpenAgentSkill.