Use when building, reviewing, or debugging a LangGraph flow that must prove a resumed run preserved its recorded semantic transitions and side-effect discipline.
Scanned 9/3/2026
Install to Claude Code
npx -y skills add zrk222/code-factory --skill langgraph-proof --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Langgraph Proof?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/zrk222-langgraph-proof)More formats (shields.io, HTML) on the badges page.
---
name: langgraph-proof
description: Use when building, reviewing, or debugging a LangGraph flow that must prove a resumed run preserved its recorded semantic transitions and side-effect discipline.
---
# Code Factory LangGraph Proof
Use Code Factory as a local proof layer around a team-owned LangGraph test
harness. It is not a graph runtime, checkpoint store, repair agent, or release
authority.
## Start with the boundary
Before relying on the MCP server, check that the current workspace has the
Code Factory CLI and inspect its stated local boundary:
```bash
factory --version
factory mcp status --root .
```
If the CLI is missing, explain that the user must install
`factoryline-code-factory>=0.43.0` in the environment that will host the local
MCP process. Do not install Python packages, change a client configuration, or
start a server without the user's approval.
## What to prove
For a LangGraph flow with a durable checkpoint or interruption path, ask for:
1. A team-owned reference harness run.
2. A separately captured resumed or interrupted run.
3. Two sealed, workspace-relative transition lineages. The recorder retains
SHA-256 facts about state and side-effect identifiers, not raw state,
prompts, or secrets.
Compare existing lineages without writing a file:
```bash
factory langgraph replay-verify \
--root . \
--reference .factory/langgraph/reference.json \
--resumed .factory/langgraph/resumed.json \
--json
```
Only add `--out .factory/langgraph/assurance.json` after the user explicitly
approves writing a local assurance receipt.
## Interpret the result precisely
- `LANGGRAPH_RESUME_PARITY_VERIFIED` means the supplied sealed lineages match
semantically and the deterministic analysis found no supported anomaly.
- `LANGGRAPH_REPLAY_DIVERGENCE` means the supplied evidence differs or exposes
duplicate completed effects, stale reads/writes, or unsafe parallel writes.
Present the incident capsule and recovery cone for human review.
- `LANGGRAPH_INPUT_REJECTED` means the comparison was rejected before a
receipt was written; fix the supplied inputs rather than guessing.
Never claim that a parity receipt proves production resilience, complete
instrumentation, external idempotency, savings, quality, or correctness beyond
the supplied lineages.
## MCP use
`factory.langgraph_assurance` is read-only. It may compare two existing
workspace-relative receipts; it cannot invoke a graph, mutate checkpoints,
replay effects, write a receipt, approve a repair, merge, deploy, publish,
send a message, or access credentials.
If a divergence needs code changes, prepare a narrow repair proposal and tests
for a human to approve. Do not authorize or execute repairs from the incident
capsule.

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!