Skip to content

feat(sleep): per-night evidence chain + live prompt registry#151

Merged
Yif-Yang merged 2 commits into
microsoft:mainfrom
Alphaxalchemy:feat/sleep-evidence-chain
Jul 21, 2026
Merged

feat(sleep): per-night evidence chain + live prompt registry#151
Yif-Yang merged 2 commits into
microsoft:mainfrom
Alphaxalchemy:feat/sleep-evidence-chain

Conversation

@Alphaxalchemy

@Alphaxalchemy Alphaxalchemy commented Jul 19, 2026

Copy link
Copy Markdown
Contributor

Per-night evidence chain + live prompt registry for SkillOpt-Sleep

Why (origin of this change)

This started from a user-side audit of what a completed sleep night actually lets you verify. Running real nights (DeepSeek via the openai_compatible backend) against real transcripts, the report tells you that the miner processed N conversations and what rules the reflect step added — but the evidential chain in between is invisible:

your transcripts → these specific mined checks → these specific replay failures → these specific edits → this gate decision

None of the intermediate artifacts (the miner's verbatim exchange, which checks each task carried, which replay attempt failed which check, the reflect model's raw reply before parsing, the gate's arithmetic) were recorded anywhere. Given how strict/fragile the acceptance gate is by design (strict improvement or nothing), a rejected night was essentially unexplainable after the fact.

The motivation goes beyond auditability. Once every miner/reflect exchange is persisted per night, three things become possible that aren't today:

  1. Check validation — programmatically ask "did the miner's checks actually correspond to what the user accepted?" and score the miner itself.
  2. Prompt regression testing — when someone edits the miner prompt, replay archived sessions through both versions and diff the mined tasks.
  3. A meta-training loop — the miner and reflect prompts are themselves just documents; SkillOpt's own gate machinery can evolve them against the archived exchanges. The system optimizing its own optimizer is exactly what this architecture supports, and this PR lays the substrate for it.

What

  • skillopt_sleep/evidence.py — append-only, thread-safe, secret-redacted evidence.jsonl written into each night's staging dir. The full chain is logged: harvested sessions → miner exchanges (verbatim prompt/reply) → mined tasks with their checks → train/val split assignment → every replay model call (phase-tagged; cache hits marked key-only) → per-task scores with failing checks named → reflect exchanges + parsed edits → gate baseline/trials and the final decision with its score arithmetic spelled out → staged artifacts. Config-gated: evidence_log (default on), evidence_max_chars truncation cap.
  • skillopt_sleep/prompts.py — central registry of the four LLM prompt templates (miner / attempt / judge / reflect). Defaults are byte-identical to the previously inlined strings (covered by a test). User overrides in prompts.json take effect on the next model call (mtime-checked) — no restart. This is what makes the prompts first-class, versionable, and eventually gate-optimizable objects.
  • cycle.py — builds dual backends from config (optimizer_* / target_*), pre-creates the staging dir so evidence lands beside the report; staging.new_staging_dir() de-collides same-second runs; latest_staging() skips non-adoptable (evidence-only) folders so a no-tasks night can never be accidentally adopted.

Tests

tests/test_sleep_evidence.py — 8 pure-stdlib, deterministic, no-network tests: chain completeness end-to-end on the mock backend, redaction/truncation/ordering, corrupt-line tolerance, evidence_log=False disable path, prompt-default byte-parity, override round-trip + live effect without restart, and the no-tasks-night adoption guard.

Full suite: no new failures vs pristine main (same 11 pre-existing research-suite loader errors on both; this branch adds 8 tests, all passing).

Live smoke: a real DeepSeek night through the openai_compatible backend produced a 95-event chain in which the gate's rejection is fully reconstructable from the log, including its formula (baseline = 0.5·0.000 + 0.5·0.333 = 0.167; candidate = 0.167 → no strict improvement → reject).

Series

This is part 1 of a 3-PR series: this PR (evidence chain + prompt registry, no UI) → #152 (a stdlib-only local dashboard that renders the chain and edits the prompts) → #153 (a new harvest source, per the "more sources" direction of #97).

Alphaxalchemy and others added 2 commits July 21, 2026 17:15
* skillopt_sleep/evidence.py — append-only, thread-safe, redacted
  evidence.jsonl per night: harvest sessions -> miner exchanges (verbatim
  prompt/reply) -> mined tasks with checks -> split assignment -> every
  replay attempt (phase-tagged, cache hits marked) -> per-task scores with
  failing checks named -> reflect exchanges + parsed edits -> gate trials
  and the final decision with its score arithmetic -> staged artifacts.
  Config-gated (evidence_log, default on; evidence_max_chars cap).

* skillopt_sleep/prompts.py — central registry of the four LLM prompt
  templates (miner/attempt/judge/reflect), byte-identical defaults to the
  previously inlined strings; user overrides in prompts.json take effect
  on the next model call (mtime-checked), no restart needed.

* cycle.py builds dual backends from config (optimizer_*/target_*),
  pre-creates the staging dir so evidence lands beside the report;
  staging.new_staging_dir() de-collides same-second runs;
  latest_staging() now skips non-adoptable (evidence-only) folders.

* tests/test_sleep_evidence.py — 8 no-network stdlib tests: chain
  completeness, redaction/truncation/ordering, disable flag, prompt
  override round-trip + live effect, no-tasks-night adoption guard.
Copilot AI review requested due to automatic review settings July 21, 2026 17:26
@Yif-Yang
Yif-Yang force-pushed the feat/sleep-evidence-chain branch from 7829552 to b12b54c Compare July 21, 2026 17:26
@Yif-Yang
Yif-Yang merged commit 38697fe into microsoft:main Jul 21, 2026
1 check passed

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds a per-night, append-only evidentiary chain (evidence.jsonl) to make SkillOpt-Sleep nights fully auditable, and introduces a centralized prompt registry with live, file-based overrides so miner/attempt/judge/reflect templates become versionable and tunable without code changes.

Changes:

  • Introduce EvidenceLog and wire evidence events across harvest → mine → replay → reflect/gate → staging.
  • Add skillopt_sleep.prompts prompt registry and refactor backend/miner to render prompts from the registry (with live overrides).
  • Improve staging lifecycle (pre-created staging dir, de-collide naming, and make latest_staging() skip non-adoptable evidence-only nights).

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 6 comments.

Show a summary per file
File Description
tests/test_sleep_evidence.py Adds unit tests for evidence logging and prompt override behavior.
tests/test_sleep_engine.py Updates a mock patch target to the new build_backend entrypoint.
skillopt_sleep/staging.py Adds unique staging-dir helper and prevents adopting evidence-only folders.
skillopt_sleep/replay.py Logs per-task replay results into evidence chain (phase-tagged).
skillopt_sleep/prompts.py New live prompt registry with defaults + override file (mtime-checked).
skillopt_sleep/llm_miner.py Moves miner prompt into registry and logs miner exchanges/tasks into evidence.
skillopt_sleep/evidence.py New thread-safe JSONL evidence logger + reader + backend attachment helpers.
skillopt_sleep/cycle.py Creates evidence log per night, pre-creates staging dir, dual-backend support wiring.
skillopt_sleep/consolidate.py Adds evidence phase tagging and logs gate baseline/trials/decision formula.
skillopt_sleep/config.py Adds config knobs for evidence logging and dual-backend split.
skillopt_sleep/backend.py Uses prompt registry for attempt/judge/reflect and logs cached/model calls + reflect exchanges.
docs/sleep/README.md Documents evidence logging behavior and retention/disable guidance.
CHANGELOG.md Notes new evidence chains and prompt registry in Unreleased additions.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread skillopt_sleep/staging.py
Comment on lines +83 to +90
def new_staging_dir(project: str) -> str:
"""A staging path that is unique even for two runs in the same second."""
base = os.path.join(staging_root(project), _ts_dir())
out, i = base, 2
while os.path.exists(out):
out = f"{base}-{i}"
i += 1
return out
Comment thread skillopt_sleep/cycle.py
Comment on lines +166 to +170
ev = EvidenceLog(
ev_path,
max_chars=int(cfg.get("evidence_max_chars", 4000) or 4000),
redact=bool(cfg.get("redact_secrets", True)),
)
Comment thread skillopt_sleep/cycle.py
Comment on lines +161 to +162
ev_path = os.path.join(
cfg.state_dir, "evidence", f"dryrun-{_ts_dir()}.jsonl")
Comment on lines +117 to +123
def _run(self, **cfg_extra):
proj = tempfile.mkdtemp()
home = tempfile.mkdtemp()
cfg = load_config(
invoked_project=proj, projects="invoked", backend="mock",
claude_home=os.path.join(home, ".claude"), auto_adopt=False,
**cfg_extra)
Comment on lines +196 to +200
proj = tempfile.mkdtemp()
home = tempfile.mkdtemp()
cfg = load_config(
invoked_project=proj, projects="invoked", backend="mock",
claude_home=os.path.join(home, ".claude"))
Comment thread skillopt_sleep/prompts.py
Comment on lines +194 to +196
os.makedirs(os.path.dirname(path) or ".", exist_ok=True)
with open(path, "w", encoding="utf-8") as f:
json.dump(current, f, ensure_ascii=False, indent=2)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants