feat(selflearn): dedicated /evolve panel with model-decided Auto trigger#311
Open
yanmxa wants to merge 1 commit into
Open
feat(selflearn): dedicated /evolve panel with model-decided Auto trigger#311yanmxa wants to merge 1 commit into
yanmxa wants to merge 1 commit into
Conversation
|
[DCO NOTIFIER] Hi @yanmxa! Some commits in this PR are missing a Signed-off-by. What to do: Or to sign off the most recent commit without changing the message: Once all commits are signed, the |
|
[REVIEW NOTIFIER] This PR requires a review from @hchenxa. Once this PR has been reviewed and has the |
|
[REBASE NOTIFIER] Hi @yanmxa! This PR has merge conflicts. The How to fix thisresolve the conflicts — Fix the merge conflicts locally, then push the resolved changes. After fixing:
|
yanmxa
force-pushed
the
enhance-self-learning
branch
from
6382cb2 to
e83ba84
Compare
yanmxa
force-pushed
the
enhance-self-learning
branch
from
e83ba84 to
05d39fe
Compare
Give self-learning its own /evolve overlay, styled like the AutoPilot panel, and drive it entirely by the model rather than by rules/cadence. Trigger: one Evolve tool is injected into the main agent when self-learning is active; calling it queues a background review at turn end. There is no cadence — the model decides when a turn is worth learning from. The tool's description is built from the enabled capabilities (tool.SelfLearnToolInfo), so the model is only invited to flag learnings the review can act on (memory off never mentions memory; all-off omits the tool entirely). Review: a background fork reflects on the turn with only skill_manage + memory_write and writes memory and/or a skill (or nothing). Skills are scoped by skill use (skill-used → update/delete, skill-free → create); the create/update/ delete gates are pure permissions. Writes go to San's own agent-memory store, kept separate from user-authored SAN.md / CLAUDE.md. Panel: one unified panel (no Skills/Memory tabs) — a shared Strategy editor, the skill permission gates, the memory store settings (enable / max KB / path), and a "Learned" drill-in that browses & prunes both agent-created skills AND agent-written memory files (grouped, view/delete each), plus a RECENT recap of both arms. The reviewer prompt keeps a shared skeleton with per-arm sections and one shared Strategy override. Docs: internal/selflearn package doc updated for the model-decided flow and manual testing steps. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
yanmxa
force-pushed
the
enhance-self-learning
branch
from
05d39fe to
f6586b3
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
A dedicated
/evolveoverlay for self-learning (Skills + Memory tabs), split out of/configand styled like the AutoPilot panel — plus a new Auto (model-decided) trigger.Skills tab
SKILL.md/ prune) with a RECENT activity recap.Evolvetool into the main agent. Calling it queues a self-learning review at turn end — scoped by skill use, bounded by the permission gates. Rule triggers are bypassed; toggling Auto rebuilds the agent so the tool appears/disappears.Memory tab
Gated to a Coming soon placeholder until the arm ships.
Engine
Evolveis a safe (auto-permitted) signal tool — the L1 review still runs out-of-band in the background fork; the model never writes skills itself.Testing
go build ./...,go vet ./internal/...,go test ./...all green; layercheck clean.Evolvetool gating (main-agent-only), panel Auto toggle, safe-tool classification.docs/packages/2-feature/selflearn.md.🤖 Generated with Claude Code