Skip to content
Back to skills

Hyperbrief Revisit

ASecurity

Invoke when a stored Hyperbrief decision's revisit_date is reached, OR when an assumed_invariant in decision_lineage is violated by recent A2A signals, OR at session-start / Stop hook / 6h cron tick to scan the ledger. Loads the original IR, prompts user for actual outcome, computes outcome-quality vs decision-quality delta (Brier score increment), appends retrospective to ledger, supersedes/affirms decision_id chain. Closes the learning loop that fire-and-forget briefs leave open. v0.6: awar...

  • 8 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 29, 2026
ai-agentsgogitapi

Works with

  • api

Security analysis

A100/100

Scanned September 29, 2026

npx -y skills add SoliEstre/EstreGenesis --skill hyperbrief-revisit --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Hyperbrief Revisit?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Hyperbrief Revisit
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/soliestre-hyperbrief-revisit/badge)](https://www.skillsdirectory.com/skills/soliestre-hyperbrief-revisit)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: hyperbrief-revisit
version: 0.6.0
description: "Invoke when a stored Hyperbrief decision's revisit_date is reached, OR when an assumed_invariant in decision_lineage is violated by recent A2A signals, OR at session-start / Stop hook / 6h cron tick to scan the ledger. Loads the original IR, prompts user for actual outcome, computes outcome-quality vs decision-quality delta (Brier score increment), appends retrospective to ledger, supersedes/affirms decision_id chain. Closes the learning loop that fire-and-forget briefs leave open. v0.6: aware of 4 new schema slots (evaluation_lenses / recommended_methodology / maturity_anchor / term_pairing) — back-compat with v0.5.6 IRs preserved."
---

# Hyperbrief Revisit — close the learning loop

You are scanning the decision ledger for decisions whose `revisit_date` has arrived (or whose `assumed_invariants` have been violated), OR a specific `decision_id` has been handed to you for review.

## 0. Schema version awareness (v0.6.0)

본 skill 은 schema v0.6.0 인지. v0.5.6 → v0.6.0 변경 사항:

- **v0.5.6 IR 도 v0.6 schema 에 valid 유지** (back-compat). 모든 신규 슬롯은 optional — 기존 ledger 항목 마이그레이션 불요.
- **신규 슬롯 4 종** (revisit 시점에 비어있을 수 있음 — 그 자체로 흠결 아님):
  1. `section_0_decision_header.evaluation_lenses[]` — 같은 결정에 대한 multi-lens 점수 (예: Lens A 모듈 GA / Lens B host marketplace).
  2. `section_8_recommendation.recommended_methodology[]` — 결정에 사용된 평가 방법론 박제 (예: `hyperbrief-v1-readiness-rubric`).
  3. `maturity_anchor` (top-level, optional) — 자의적 라벨 (v1.0/GA/stable) 에 측정 가능한 anchored 기준.
  4. `AudienceProfileFallback.term_pairing` — tone 프로파일의 표시 정책 + 적용 범위 (mode E/I/N · scope C/D/B/R · retroactive_apply).
- **v0.5.6 caller 도 v0.6 skill 호환**: 이전 버전 IR 를 그대로 입력해도 본 skill 동작. 신규 슬롯 부재 시 retrospective 질문에서 "옵션으로 채우기" 만 제시 (강제 아님).

## 1. Trigger conditions

Run this skill if ANY of these apply:

- Session-start / Stop hook / 6h cron tick → scan ledger for `revisit_date <= now AND outcome_actual == null`.
- Inbound A2A signal contradicts an `assumed_invariant` in any active decision's lineage.
- User explicitly invokes `/hyperbrief revisit <decision_id>`.
- `hyperbrief` skill's parent-decision check found a stale parent.
- **v0.6 추가**: `maturity_anchor.claimed_label` 이 부착된 결정의 경우, anchor methodology 의 current_score 가 threshold 를 (재)교차했는지 감지된 시점.

## 2. Ledger scan

Locate the ledger at one of:

- Standalone: `.agent/_decisions/*.json` (one file per decision_id).
- Constellation-integrated: Constellation `state.json` → `decisions[]` array, OR SSE endpoint `/api/decisions/stream`.

For each entry where `revisit_date <= now AND outcome_actual == null`:

1. Load the original IR (full JSON). **Schema version 확인** — v0.5.6 / v0.6.0 둘 다 처리. 누락 슬롯은 정상 (back-compat).
2. Surface to the user as a `HyperbriefRevisit` card (or inline text in standalone mode).
3. Block other low-priority work until the user responds (revisit > new work).

## 3. The 8-question retrospective (+ v0.6 옵션 4 질문)

Ask the user, ONE question at a time (avoid framing-effect batching):

1. **Outcome** — "이 결정의 실제 결과는 어땠어요? 1-2 문장으로." (Free-form text → `outcome_actual`.)
2. **Confidence delta** — "결정 당시 confidence 는 `<original_confidence.point_estimate>` 였어요. 지금 돌아보면 적절했어요? (under-confident / well-calibrated / over-confident)"
3. **Falsifier check** — "결정에 부착된 falsification trigger (`<what_to_observe>` at `<when>`, threshold `<threshold>`) 는 trip 됐어요? (yes / no / not yet measurable)"
4. **Assumed-invariant check** — "결정 당시 가정 (`<assumed_invariants[]>`) 은 여전히 유효해요? 위반된 것이 있다면 어떤 것?"
5. **Pre-mortem hit** — "결정 당시 pre-mortem 시나리오 (`<pre_mortem_scenarios[].failure_path>`) 중 실제로 발생한 것이 있어요?"
6. **Defeater check** — "결정 당시 defeaters (`<defeaters[]>`) 중 fire 된 것이 있어요?"
7. **Decision quality** — "결과의 좋고 나쁨과 별개로, *그 시점에 그 정보로* 결정 자체는 합리적이었어요? (Y/N + 이유)"
8. **Successor decision** — "이 결정을 supersede 하거나 affirm 하는 후속 결정이 필요해요? (yes — open new hyperbrief / no — affirm / partial — defer to next checkpoint)"

### 3.1 v0.6 옵션 질문 (신규 슬롯이 비어있고 사용자가 채우길 원하는 경우만)

기존 결정이 v0.5.6 IR 이라 4 신규 슬롯 중 일부가 비어있을 때, retrospective 끝에 1-line 옵션으로 제시 ("필요하면 채울 수 있어요 / 건너뛰어도 돼요"):

9. **evaluation_lenses** (slot empty 시) — "이 결정을 multi-lens 로 재평가해볼 lens (관점) 가 있어요? (예: 모듈 GA lens vs host marketplace lens) 없으면 skip."
10. **recommended_methodology** (slot empty 시) — "이 결정에 사용한 평가 방법론 (rubric/framework) 을 박제할 만한 것이 있어요? (예: readiness rubric §11.5 같은 메타-결정 도구) 없으면 skip."
11. **maturity_anchor** (slot empty + 결정이 라벨링 결정 시) — "이 결정이 'v1.0/GA/stable' 같은 자의적 maturity 라벨을 부착했다면, anchor methodology + threshold 명시해볼래요? 라벨링 결정 아니면 skip."
12. **term_pairing** (AudienceProfile slot empty 시) — "tone 프로파일에 term_pairing (약어 병기 정책: mode E/I/N · scope C/D/B/R) 추가할래요? 기본 N/[D] 로 두려면 skip."

**규율**: 옵션 질문은 skip 이 기본. 사용자가 명시적으로 채우길 원할 때만 진행. 빈 슬롯 자체는 흠결 신호 아님 (v0.5.6 vintage 결정의 정상 상태).

## 4. Brier increment computation

For each `toulmin_predictions[]` and `§8.confidence` that has a `point_estimate`:

```
brier_increment = (actual_outcome_binary - point_estimate)^2
```

Append the increment + cumulative average to the ledger entry:

```jsonc
{ "decision_id": "hb-...",
  "schema_version_at_decision": "0.5.6" | "0.6.0",
  "schema_version_at_revisit": "0.6.0",
  "outcome_actual": "<user text>",
  "decision_quality": "reasonable" | "premature" | "biased" | "lucky",
  "outcome_quality_vs_decision_quality_delta": "outcome-better-than-decision" | "aligned" | "outcome-worse-than-decision",
  "brier_increment": { "predictions": [...], "cumulative_average": <float> },
  "assumed_invariants_status": { "<inv1>": "still valid" | "violated", ... },
  "successor_action": "new_brief:<new_id>" | "affirmed" | "deferred:<new_date>",
  "v06_slots_filled": { "evaluation_lenses": <bool>, "recommended_methodology": <bool>, "maturity_anchor": <bool>, "term_pairing": <bool> } }
```

### 4.1 v0.6 추가 retrospective 지표 (해당 슬롯이 채워진 결정 한정)

- **`evaluation_lenses` 가 있던 결정**: 각 lens 의 verdict (below/at/above) 가 실제 결과와 정렬됐는지 lens-by-lens 평가. 일치 lens / 불일치 lens 카운트 retrospective 에 기록.
- **`recommended_methodology` 가 있던 결정**: 박제된 방법론이 retrospective 시점에도 여전히 적용 가능한지 (`applicability` 검증). 방법론 자체의 수명도 추적.
- **`maturity_anchor` 가 있던 결정**: `current_score` vs `threshold` 의 gap 이 결정 당시 vs revisit 시점 사이에 어떻게 변했는지 (gap 축소 / 확대 / 유지). `verdict` (meets/near/short) 변화 기록.
- **`term_pairing` 이 있던 결정**: `retroactive_apply` 가 prompt/Y 였다면 실제 소급 적용이 surface (D/B/R) 에 일관되게 반영됐는지 점검 (C 는 자동 forward 라 점검 외).

## 5. Lineage propagation

If `successor_action == "new_brief:<new_id>"`:

- The new brief's `parent_decision_ids[]` MUST include this `decision_id`.
- This decision's `status` flips to `"superseded_by:<new_id>"`.
- **v0.6**: 후속 brief 가 v0.6 schema 로 발행되면, 부모 결정의 `evaluation_lenses` / `recommended_methodology` / `maturity_anchor` 를 inherit 가능 (skill `hyperbrief` 가 부모 IR 참조 시 신규 슬롯도 후보로 surface).

If `assumed_invariants_status` shows any violation:

- All descendant decisions (decisions whose `parent_decision_ids[]` contains this id) MUST be revisited next.
- Auto-queue them to the top of the revisit queue.

## 6. Constellation integration

If Constellation is present:

- Emit `CUSTOM/HyperbriefRevisit` card to the user-board for inline rendering.
- POST the completed retrospective to `/api/decisions/<id>/retrospective`.
- Subscribe to `/api/decisions/stream` for live invalidation events.
- **v0.6**: `term_pairing.scope` 에 `B` (board) 포함된 결정이 revisit 될 때, 보드 채널로 re-emit 신호 전송 (소급 적용 surface 갱신 트리거). `retroactive_apply == "Y"` 면 자동, `"prompt"` 면 사용자 확인 후.

## 7. Standalone mode

- Write retrospective JSON to `.agent/_decisions/<id>.retro.json`.
- Update `.agent/_decisions/<id>.json` with `status` change.
- If `successor_action == new_brief`, invoke the `hyperbrief` skill with `parent_decision_ids` pre-populated.
- **v0.6**: v0.5.6 IR 을 revisit 후 신규 슬롯 일부를 채워 갱신하는 경우, 원본 ledger 항목은 `schema_version` 만 `0.6.0` 으로 bump 하고 신규 슬롯 append. 기존 필드는 건드리지 않음 (in-place additive migration).

## 8. Anti-patterns

- **Outcome-decision quality conflation** (anti-pattern AF-17): a bad outcome from a reasonable decision is a calibration signal, NOT a decision-quality failure. Surface this distinction to the user (question 7 above is the gate).
- **Hindsight-bias retrofit**: do NOT let the user re-grade the decision higher just because the outcome was good (or lower because bad). Question 7's framing ("그 시점에 그 정보로") is the debias.
- **Stale revisit ignore**: if `revisit_date - now > 30d` (overdue more than 30 days), surface it as a `[stale-revisit]` flag, do not silently skip.
- **v0.6 슬롯 강제 채움**: v0.5.6 vintage 결정의 빈 신규 슬롯을 retrofit 강제 금지. 옵션 질문은 skip-by-default. 빈 슬롯 = 흠결 NOT, 단지 그 결정이 그 차원에서 평가되지 않았다는 사실 기록일 뿐.
- **Schema-version drift silent**: revisit 시점에 schema version mismatch (decision v0.5.6 / current v0.6.0) 를 ledger 에 기록하지 않고 silent migrate 금지. `schema_version_at_decision` + `schema_version_at_revisit` 양쪽 명시.

## Reference

Full spec: https://github.com/SoliEstre/EstreGenesis/blob/main/Hyperbrief.md (§9 Decision Capture + §5 Epistemic Discipline → Brier loop)
v0.6 change log: Hyperbrief.md frontmatter + CHANGELOG.md entry for v0.6.0 minor bump (additive 4 slots, back-compat).

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.