A
Project memory
Decisions, facts, procedures, and warnings that belong to the project you are working on.
~/.chat2skill/c2s.db
Local-first memory for coding agents
Make every session compound.
Chat2Skill is an open-source local memory and reusable skills plugin for AI coding agents. It learns the decisions, preferences, and project context hidden in your conversations, then brings the right pieces back before the next task begins.
Skills and history stay on your machine. Extraction is stateless.
CHAT2SKILL / PROJECT MEMORY
local
>_
refactor storage without changing the contract
RETRIEVED FOR THIS TASK3 items
Keep local data durableProject memory / storage
0.94
Preserve the current CLI pathPreference / workflow
0.88
Run focused tests before refactoringProcedure / verification
0.81
Your context, ready before the work starts.
LOCAL-FIRSTMemory stays in your local store
FEEDBACK-LEDCorrections become reusable rules
MODEL-AGNOSTICBring the provider you already use
MIT LICENSEDOpen source and inspectable
01 / The learning loop
Memory is a workflow, not a scrapbook.
Chat2Skill watches for signals that should change future work. It distills them into durable artifacts, validates what is worth keeping, and retrieves only what matches the task in front of you.
SESSION END signal detected
A correction is more than a correction.
When you say "keep this behavior" or "that is not verified," Chat2Skill recognizes a durable signal instead of letting it disappear with the transcript.
01signal.typeconstraint
02evidenceuser correction
03nextanalyze -> validate -> store
02 / What it keeps
Three layers. One quieter next session.
Every artifact has a job, a scope, and a place on your machine.
B
Atomized skills
Focused, reusable SKILL.md files for preferences, procedures, and successful patterns.
skills/<user>/<name>/
C
Response guard
Evidence-based rules that make the assistant distinguish verified conclusions from missing evidence.
response_guard
03 / Admin Console
See what your assistant learned. Shape what happens next.
Chat2Skill includes a local-only management page for reviewing project memory and skills. Search the evidence, edit what needs correction, and rebuild the project skill before the next session.
skillsmemoriespromptsproject-skilloverviewevals
response-guard-evidenceactive · preference · evidence 4updated today
run-focused-testsactive · procedure · evidence 3updated yesterday
preserve-cli-contractactive · constraint · evidence 2updated Aug 21
SKILL DETAIL
response-guard-evidence
Typepreference
Statusactive
Content
Use definitive wording for verified facts. Name the missing evidence and next validation step when evidence is absent.
SaveMark Archived
Review by project
Switch between projects discovered in your local database and see active skills, memories, and update times at a glance.
Edit the source of truth
Search, edit, activate, archive, or delete atomized skills and project memories instead of accepting every extraction blindly.
Rebuild the project skill
Inspect source snapshots, edit the generated PROJECT_SKILL.md, rebuild it, and run evaluations from one local workspace.
Start locally
Read the Admin Console guide
python3 scripts/chat2skill_admin.py04 / Integrations
Bring it to the agent you already use.
Native lifecycle support is available where the host exposes hooks. Other agents can use the same retrieval and learning scripts through project rules or the CLI.
Native lifecycle integrationsretrieve + learn + guard where exposed
CClaude CodeNative plugin marketplace
⌘CodexNative plugin and hooks
DDeepSeek HarnessNative Cordis bundle
↗CursorNative plugin and project rule
Rules and CLI pathsretrieval or manual learning
OpenCodeGitHub CopilotWindsurf / CascadeKiroClineAider / generic agents
05 / Start here
Use the host you already trust.
Install the plugin, let the first hook initialize local storage, and keep your existing model credentials in your own environment.
Python 3.10+
Bring your own key
Open source
Install from the marketplace
codex plugin marketplace add rxa3c/chat2skill
codex06 / Privacy by architecture
Your context should work for you, not leave your machine by default.
01
Local retrieval
Prompt retrieval reads your local SQLite store. It does not call the cloud.
02
Stateless extraction
Stop-hook transcripts are analyzed in memory and are not persisted server-side.
03
Credential ownership
Use your own API key or approved token. The Chat2Skill server never persists provider credentials.