jev_project_context
Evidence-first experiment-memory skill for coding agents, with an optional Jev audit layer that checks provenance and claim support.
Overview
Project Context V2 (repository jev_project_context) is a skill that turns a coding agent such as Claude Code into a disciplined lab-notebook keeper for research projects. It provides ten operations across a per-session loop (start with task-conditioned context, frame questions and falsifiers, plan with stable experiment IDs and frozen protocols, record with a provenance gate, end with a journal handoff), plus synthesis of traceable observations into facts, correction (freeze, invalidate, supersede) and read-only audits. A controlled validity vocabulary keeps claims traceable.
Jev appears in an optional audit layer. After a free local regex check for structural problems, jev_doctor.py makes one batched Jev call asking whether provenance is recoverable, whether headlines match tables, whether observations are clean and whether claims are supported. A confidence gate passes silently at 0.5 or above, raises a warning for high-confidence findings, and sends low-confidence ones to manual review. Chinese documentation is also provided.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is the provenance for this record recoverable? | yes · no |
| Yes / no | Does the headline match the table? | yes · no |
| Yes / no | Is this claim supported by the recorded evidence? | yes · no |
How it works
- Adopt the skill and initialise project memory
- Frame, plan and record experiments each session
- Run the local structural doctor
- Optionally run one batched Jev audit
- Act on warnings or manual-review items
Strengths
- Keeps research claims traceable to evidence
- Jev layer is optional and batched
- Confidence-gated audit output
Caveats
- Jev audit adds a hosted dependency
- Discipline depends on the agent following the skill
At a glance
- Type
- Agent skill
- Status
- beta
- Language
- Python
- Runtime
- Agent skill with helper scripts
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Evidence-first long-term experiment memory skill for AI coding agents, with optional Jev decision-model layers
logicrw/awesome-jev-projectsEvidence-first long-term experiment memory skill for AI coding agents, with optional Jev decision-model layers.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev check?
Whether provenance is recoverable, headlines match tables, observations are clean and claims are supported.
Is Jev required?
No. It is an optional audit layer on top of local checks.
Which agents work with it?
Claude Code, ZCode and other skills-compatible agents.
Related projects
Local memory for coding agents that stores verbatim quotes and has Jev grade memory lines as keep, stale, wrong or unsupported.
ProjectbetaPythonEvidence gate for RAG: a System One model judges whether retrieved excerpts support an answer before the LLM is allowed to speak.
MCP serverbetaPythonLocal long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
ProjectproductionTypeScriptVideo summarizer that ties every line to its second; Jev checks each summary sentence against what was said, with numbers checked in code.
ProjectbetaPythonMCP server with record and recall tools: facts are logged during a session and Jev keeps only those relevant to the next task.
MCP serverbetaTypeScriptSmall stdio MCP server with ask, rank and verify tools backed by Jev, a local event log and installers for several agent harnesses.
MCP serverbetaTypeScript