# Deterministic Packaging

> Deterministic Packaging — Skill 90

- Skill: `nickgallick/deterministic-packaging` (Agent Skill)
- Install (CLI): `npx skillmds@latest add nickgallick/deterministic-packaging`
- Raw SKILL.md: https://api.skillmd.com/api/skills/nickgallick/deterministic-packaging/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: nickgallick (https://skillmd.com/u/nickgallick)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/nickgallick/deterministic-packaging

---

# Deterministic Packaging — Skill 90

## Purpose
Ensure challenges are reproducible and auditable. Same template + same seed + same mutation parameters = identical challenge.

## Reproducibility Requirements

Every challenge instance must be reproducible from its seed:

| Input | Description |
|-------|-------------|
| template_id | Which template was used |
| template_version | Exact version of the template |
| generation_seed | 8-char hex seed |
| mutation_params | Full mutation parameters object |
| generation_timestamp | When it was generated |
| gauntlet_version | Which version of Gauntlet generated it |

**Same inputs = identical outputs.** This enables: re-running calibration, auditing disputed scores, reproducing bugs, verifying tampering.

## Audit Trail Structure

```json
{
  "audit": {
    "instance_id": "BOUTS-2026-XXXX",
    "template_id": "tmpl-{family}-v{N}",
    "template_version": "integer",
    "generation_seed": "8-char hex",
    "mutation_params": {
      "framework": "string",
      "database": "string",
      "bug_types": ["string"],
      "red_herring_style": "string",
      "log_noise_level": "string",
      "mutations_applied": ["mutation_type"]
    },
    "generated_by": "gauntlet-v{version}",
    "generated_at": "ISO-8601",
    "calibration_run_id": "string",
    "calibration_result": "passed | failed",
    "published_at": "ISO-8601 | null",
    "retired_at": "ISO-8601 | null",
    "quarantined_at": "ISO-8601 | null",
    "content_hash": "sha256:..."
  }
}
```

## Content Hashing

1. **At generation:** Hash the entire challenge package (workspace files + tests + rubrics)
2. **Store the hash** in the audit trail
3. **Before each run:** Verify hash matches (ensures nobody tampered post-publication)
4. **If hash mismatch:** Quarantine immediately, alert ops

## Version Pinning

- Record which version of Gauntlet generated each challenge
- If Gauntlet is updated, old challenges retain their original generation metadata
- New calibration runs after a Gauntlet update should verify old challenges still calibrate correctly
- If calibration drifts after update → investigate whether Gauntlet changes affected challenge quality

## Audit Queries

The audit trail must support:

| Query | Purpose |
|-------|---------|
| "Show all instances from template X" | Template health analysis |
| "Show all instances generated by Gauntlet version Y" | Version impact analysis |
| "Show the full lineage of instance Z" | Dispute investigation |
| "Verify instance Z hasn't been modified" | Tamper detection |
| "Reproduce instance Z from seed" | Calibration re-run |

## Integration Points

- **Structured Output** (Skill 77): Audit trail wraps the challenge JSON
- **Calibration Packaging** (Skill 81): Lineage stored in meta/lineage.json
- **Challenge Genealogy** (Skill 84): Audit data feeds genealogy analysis
- **Defensibility Reporting** (Skill 57): Audit trail is core defensibility evidence