# Ais Flows Ready Gate

> Use before releasing a product, feature, release candidate, or handoff when an AI agent must separate facts from assumptions, check evidence depth, block shallow ready claims, and produce a concrete fix feed.

- Skill: `aisflows/ais-flows-ready-gate` (Agent Skill, multi-file: 9 files)
- Install (CLI): `npx skillmds@latest add aisflows/ais-flows-ready-gate`
- Raw SKILL.md: https://api.skillmd.com/api/skills/aisflows/ais-flows-ready-gate/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- License: permissioned
- Author: aisflows (https://skillmd.com/u/aisflows)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/aisflows/ais-flows-ready-gate

---


# AIS FLOWS / Ready Gate

Use this skill to prevent shallow product audits, unsafe release claims, and agent work that looks complete without enough evidence.

Default role: audit and report. Do not modify the product unless the user explicitly asks for implementation.

## Choose The Lightest Gate

Use `DEVELOPMENT_STAGE_PRODUCT_GATE` when the product is still under active development.

Use `PRE_RELEASE_PRODUCT_GATE` when real users, auth, storage, billing, integrations, production data, packages, installers, or public claims are involved.

Use `PRE_RELEASE_MEDIA_PRODUCT_GATE` for media-heavy products with generation, assets, timeline, render, export, storage, or creative workflows.

Use `DEEP_PRODUCT_AUDIT` only when the user expects a serious audit of a large product, not a quick report review.

## Core Rules

1. Separate current development baseline from release readiness.
2. Do not turn future release blockers into current-baseline failures unless they block the next planned phase.
3. Do not claim runtime, storage, billing, export, provider, package, or security facts without evidence.
4. Do not treat private runtime material as publicly exposed unless evidence proves exposure.
5. A command timeout is not automatically a product failure. If the runner may be invalid, mark `HARNESS_RESULT_INVALID`.
6. A change can fail even when behavior works if it moves complexity to the wrong owner, creates thin wrappers, leaks logic across layers, or makes the codebase harder to maintain.
7. Send product teams a short verdict plus fix feed by default, not a full audit dump.

## Weak-Model Self-Containment

The user must not need to know a special low-model prompt.

For every non-trivial audit, build your own audit charter:

```text
target product:
stage:
available evidence:
selected gates:
forbidden actions:
missing evidence:
```

For every finding, preserve:

```text
Observed:
Constraint:
Inference:
Verdict:
Evidence gap:
```

If you cannot keep this structure, return `REWORK_WEAK_MODEL_GUARD_FAILED`.

## Large Product Depth

Before a verdict on a large product, state:

```text
audit depth: level 1 / level 2 / level 3
runtime freshness: freshly run / saved evidence only / not available
coverage map:
evidence counter:
uninspected high-risk areas:
```

Verdict limits:

- no source sampling -> level 1 only;
- source sampling but no fresh runtime/build/browser/API validation -> level 2 at most;
- fewer than 12 concrete evidence points for a large product -> at most `PASS_WITH_RISKS`;
- missing critical zones -> `NOT_DEEP_ENOUGH` or named `Evidence Gaps`.

## Required Output

```text
Selected gates:
Runtime evidence freshness:
Audit depth:
Coverage map:

Findings:
- Observed:
  Constraint:
  Inference:
  Verdict:
  Evidence gap:
  Owner:
  Fix:
  Validation:

Development baseline verdict:
Release verdict:
Product-facing fix feed:
Not facts without evidence:
Residual risk:
```

## Product-Facing Fix Feed

For every actionable fix, include:

```text
owner:
severity:
accepted/current-baseline impact:
release impact:
required change:
validation command or evidence path:
stale/superseded status:
```

Feed the fix feed and short verdict summary to product owners. Keep the full audit report as evidence history unless the product explicitly asks for it.

## Forbidden Without Explicit Approval

- reading `.env`, provider profiles, browser profiles, cookies, tokens, SSH keys, or secrets;
- logging into accounts;
- running product code, package scripts, installers, unknown binaries, `npx`, or provider calls;
- changing routes, billing, admin settings, publishing, or production data;
- claiming private material is public without proof.

## References

Load only what the task needs:

- `references/gate-checklists.md`
- `references/output-verdicts.md`
- `references/public-package-safety.md`