Kali-Decoder / Kali-Decoder/Midnight-Skills

Explore a dynamic layer for the fast-moving facts (docs/versions/API surface)

Open
#18 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
TypeScript
Stars
16
Forks
5
PR merge metrics
No merged PRs in 30d

Description

This one is a longer-horizon design discussion rather than a fix — feel free to treat it as a thinking-out-loud prompt.

**Observation.** The content in the repo naturally divides into two kinds:

1. **Durable reasoning** — mental models, privacy patterns, decision tables, "common mistakes." `compact/`, `testing/`, and `security/` are strong examples: they age well and are hard for an agent to reconstruct on its own. This is the repo's real differentiator.
2. **Volatile facts** — exact package versions, full SDK call signatures, complete app scaffolds. These change release-to-release, and once frozen in Markdown they can only go stale.

Static skill files are an excellent home for (1). For (2), a static snapshot is fighting the release cadence — which is what motivates other issues here.

**Why it matters.** Separating "what to reason about" from "what to look up fresh" would let each part play to its strength: the skills stay lean and durable, and the volatile facts are always current.

**Suggested direction (very open):**

- A lightweight retrieval path for the volatile layer — e.g. an MCP server (or a documented fetch step) that resolves current versions and pulls live docs/API surface on demand — so skills reference *where to get the current answer* rather than embedding a point-in-time copy.
- There's already verification- and retrieval-oriented tooling elsewhere in the Midnight agent ecosystem that could be reused rather than rebuilt; worth a look before investing in something bespoke.
- Even without an MCP server, a documented "resolve this live" convention captures much of the benefit.

Would be glad to sketch what a minimal version-resolution + docs-retrieval endpoint could look like if there's interest.

Contributor guide

Open the contributing guide

Research direction

The issue names no implementation files, tests, or entry points. Start by reviewing the existing retrieval-oriented tooling in the Midnight agent ecosystem and the repo's compact/, testing/, and security/ content; done would be an agreed minimal design for resolving current versions and documentation, or a documented live-resolution convention.

Written by the indexing model from the issue text.

Assessment

Domain
api, documentation, tooling
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.