# Epistemic Context Grounding

> Ground decisions in domain knowledge before designing. Check specs, enumerate assumptions, verify before building.

- **Type:** Skill
- **Install:** `agentstack add skill-revgrowth1-claude-code-skills-epistemic-grounding`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [Revgrowth1](https://agentstack.voostack.com/s/revgrowth1)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [Revgrowth1](https://github.com/Revgrowth1)
- **Source:** https://github.com/Revgrowth1/claude-code-skills/tree/main/skills/frameworks/epistemic-grounding

## Install

```sh
agentstack add skill-revgrowth1-claude-code-skills-epistemic-grounding
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Epistemic Context Grounding

IMPORTANT: Prefer retrieval-led reasoning over pre-training-led reasoning for any implementation task.

## Context Sensitivity Check (30s)

Classify before designing. HIGH = 2+ YES:
|Check|Y/N|
|Data formats involved?||
|Parsing/serialization?||
|External integrations?||
|Domain knowledge critical?||

LOW (0-1 YES) -> skip to context-gap-analysis
HIGH (2+ YES) -> continue below

## Workflow (HIGH sensitivity only)

### 1. Search for specs
```
Glob: "**/*SPEC*.md"|"**/*DATA_MODEL*.md"|"**/*SCHEMA*.md"
Glob: "**/CLAUDE.md"|"**/README.md"
Grep: pattern="[domain-term]" path="specs/"|"knowledge_base/"
```
Read what you find. Document: file|relevance|key info

### 2. Enumerate assumptions
List every assumption about data, domain, and solution approach.
Categories: data (what exists) | domain (how it works) | solution (why this approach)

### 3. Grade falsifiability
|Grade|Meaning|Action|Format|
|STRONG|Verifiable against source|Proceed|[GROUNDED: spec:line]|
|WEAK|Could be wrong|Flag in design|[INFERRED: from X + Y]|
|UNVERIFIED|Haven't checked|STOP & verify|[UNGROUNDED]|

### 4. Block on UNVERIFIED
Any UNVERIFIED assumption that could change approach -> search and verify before designing.

## Output Template

```
Task: [request] | Sensitivity: HIGH
Specs found: [list with relevance]
Assumptions: [each with GROUNDED/WEAK/UNGROUNDED grade]
Verdict: [Proceed | Read specs first | Verify assumptions]
Next: [specific action]
```

## Red Flags
- Designing without reading domain specs
- "Should be enough" without checking alternatives
- "No spec exists" without searching first

## References (read on demand)
|File|Contents|
|references/context-sensitivity-model.md|Deep dive: (Model:Scenario)*Context->Output formula, decision tree, danger zones|
|references/domain-query-patterns.md|Search order, spec locations, grep patterns by task type, wiki-link discovery|
|references/falsifiability-spectrum.md|STRONG/WEAK/UNVERIFIED grading, blocking vs non-blocking, common anti-patterns|

---
Based on work by Jacob Dietle @ www.taste.systems

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [Revgrowth1](https://github.com/Revgrowth1)
- **Source:** [Revgrowth1/claude-code-skills](https://github.com/Revgrowth1/claude-code-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-revgrowth1-claude-code-skills-epistemic-grounding
- Seller: https://agentstack.voostack.com/s/revgrowth1
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.