Repository navigation
feat: add the self-evolution demo - #111
Merged
al2wang merged 8 commits intoOct 12, 2026
Merged
Conversation
al2wang
added this pull request to stack #110
September 29, 2026 04:46
al2wang
force-pushed
the
alex/self-evolve-demo
branch
from
September 29, 2026 05:02
5d1b0d2 to
1307357
Compare
al2wang
force-pushed
the
alex/self-evolve-demo
branch
from
October 1, 2026 01:59
1307357 to
7c78e69
Compare
al2wang
force-pushed
the
alex/self-evolve-demo
branch
from
October 1, 2026 03:21
7c78e69 to
fdb1ac3
Compare
…kill A self-evolving system is hard to believe from a description, so this narrates the lifecycle with the evidence for each step: an episode that failed its own verifier and then passed it, the Experience distilled from that recovery and held as a candidate, its approval, the sanitized Skill published from it, two independent receipts, the gate that verified it, and a fresh agent retrieving exactly that Skill and solving a task it had never seen. It ends with the candidate distilled from poisoned text, which never leaves quarantine. Replay is deterministic and needs no provider: it reads a fixture reduced from a real run, with every minted identifier renamed and every path, host, token or key-shaped string refused rather than shipped. The fixture carries a digest the demo checks before printing. Live mode walks the same lifecycle against a running daemon and stops where the real system stops; it never reads the fixture and never fills a step in for you. AI-assisted drafting was used; the resulting changes were reviewed and verified by the author. Signed-off-by: al2wang <lightever131@gmail.com>
The demo's replay is a recorded run and stays as it was: digest-checked, offline,
byte-identical between runs. Live mode was written against the endpoints as they
stood before the Experience core and the shop landed, so it asked for
`{"items": [...]}` from unscoped paths that no longer exist.
Live mode now walks the lifecycle through the product's own scoped endpoints: the
Experiences a scope holds, candidate and approved; the Skills it can see by
status; the receipts behind the newest verified Skill, with each evaluator's
independence and origin; and how often that Skill has actually been injected into
a turn. A registry address and token add the Forum stage, reading the lineage the
Skill belongs to.
It still refuses to invent anything, and now says so precisely. There is no
default scope, because a guessed one would read someone else's state or an empty
one and call it a result. A missing approved Experience, a Skill that never
reached verified, fewer than two independent evaluators behind it - imported
receipts explicitly not counted - or a Skill nobody has reused each stop the demo
naming what is missing and how to produce it.
The live path is covered without a provider or a daemon: a local stand-in answers
the same shapes the API returns, and the tests assert the complete lifecycle is
reported from that state, that every scoped call carries the scope, and that each
of the four incomplete states stops the run. A guide describes the two modes and
the trust boundary a Skill crosses.
AI-assisted drafting was used; the resulting changes were reviewed and verified by the author.
Signed-off-by: al2wang <lightever131@gmail.com>
al2wang
force-pushed
the
alex/self-evolve-demo
branch
from
October 1, 2026 19:19
fdb1ac3 to
c4b4555
Compare
Brings #100's review, already propagated through #101 and #109, down to the demo branch. No conflict: this branch touches no daemon code, and its lock was already consistent once the upstream disambiguation of the two `sha2` versions arrived. Gate on the merged tree: `cargo fmt --all -- --check`, `cargo clippy --locked --workspace --all-targets -- -D warnings`, daemon experience (15) and skill (22) suites pass. The same three approval-resume tests fail on this host as on pristine #100, and pass in CI. AI-assisted drafting was used; the resulting changes were reviewed and verified by the author. Signed-off-by: al2wang <lightever131@gmail.com>
One page above the demo and the skill-shop guide that says why the stack exists, how its three loops fit together (local, population, evolution), where the trust boundary sits, what privacy and safety guarantees hold, the opt-in settings by their actual names, the demo path, and the early evaluation signals with their scope sentence. Linked from the docs index; docs and docs-site copies are identical. AI-assisted drafting was used; the resulting changes were reviewed and verified by the author. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Signed-off-by: al2wang <lightever131@gmail.com>
This was referenced Oct 9, 2026
… overview The overview's "Early evaluation signals" section listed development improvements without saying that the completed confirmations of the same two settings reversed. It also said applicability-aware retrieval beat unconditional and random injection without saying it stayed below the no-memory baseline. A product guide is the wrong place to carry those qualifications, so the section goes. The architecture, configuration, trust boundaries and demo instructions are unchanged. The docs-site copy is regenerated by `npm run docs:sync`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: al2wang <lightever131@gmail.com>
Signed-off-by: al2wang <lightever131@gmail.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: al2wang <lightever131@gmail.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
al2wang
merged commit Oct 12, 2026
5bca789
into
alex/verified-self-evolving-experience
12 checks passed
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
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.
Connects the complete self-evolution lifecycle into a runnable, inspectable demo.
Stacked on #109; the final user-facing slice. S-Code now implements an end-to-end
verified self-evolution stack.
What it adds
One command that shows the whole path on real product state:
scripts/demo-self-evolve.sh # deterministic replay, no provider scripts/demo-self-evolve.sh --live \ --organization ORG --team TEAM --actor ACTOR [--registry URL --registry-token TOKEN]corrective trajectory → Experience candidate → evaluation / approval → local reuse →
Skill publication → independent evaluation receipts → verified Skill → fresh-agent
retrieval → challenge / refinement → supersession.
Replay narrates a recorded run reduced to its lifecycle and stripped of anything
private; the fixture is checked against its own digest and two runs print the same bytes.
Live walks the same lifecycle against your own daemon through the product's scoped
endpoints and prints only state that run read; a stage that cannot be shown stops the demo
saying what is missing. Nothing is synthesized in either mode, and the deterministic demo
embeds no benchmark results.
Docs
Adds the top-level overview Self-Evolving S-Code (
docs/guides/self-evolving-s-code.mdand the docs-site copy): why, the three loops, trust, privacy and safety, opt-in
configuration with the actual setting names, the demo path, and early evaluation signals.
This update
Carries the current #109 head; no conflicts; the delta over #109 is byte-identical to the
original demo slice plus the overview page.
Non-goals
No runtime, crate or migration changes. No research artifacts.
This PR covers the feature implementation and its engineering validation; it does not claim improved held-out task performance.
Parent: #109.
🤖 Generated with Claude Code