一键导入
change
Log a requirement change to the BRD changelog, run impact analysis, and cascade updates through affected specs, design, and implementation.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Log a requirement change to the BRD changelog, run impact analysis, and cascade updates through affected specs, design, and implementation.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Switch model mid-session preserving thinking blocks. Used by /model command and by BRD §6.2 failover when the primary provider rate-limits or fails.
Six-phase ReAct loop (pre-check, thinking, self-critique, action, tool, post). Activates per-workflow via the thinking_level knob in config/workflows.yaml. Replaces standard ReAct for workflows where independent verification of the reasoning improves outcomes.
Mines completed sessions for repeating {tool sequence → outcome} tuples. Scores by frequency × success_rate × novelty. High-scoring tuples become "instincts" — candidate skill seeds — in instincts/pending/.
Walk-back algorithm for the Spec-Auditor subagent. Given a failure at phase N, identifies the earliest phase whose spec, if tightened, would have prevented the failure. Proposes a surgical amendment.
Sessions stored as trees (not lists). /fork creates a branch from any point; /tree navigates; /branch labels a path; /export produces HTML for review. All branches live in one session file under sessions/<project>/<session_id>.json.
Progressive context refinement for subagents. Don't dump everything; retrieve in passes. Start with a list of file paths and one-line abstracts; load full content only when the abstract proves insufficient.
| name | change |
| description | Log a requirement change to the BRD changelog, run impact analysis, and cascade updates through affected specs, design, and implementation. |
| argument-hint | [change description] |
Log a requirement change, trace its impact across the project, and cascade updates through affected specs, design, and implementation. Every change is versioned in the BRD changelog so agents can detect staleness.
If a change description was provided as an argument, use it. Otherwise, ask the user:
What changed? Describe the requirement change in one or two sentences.
Also ask:
Is this an addition, modification, or removal?
Record the change type for the changelog entry.
Read specs/brd/brd.md and identify which section(s) the change affects. Present the mapping to the user:
This change appears to affect:
- Section 3.2 — Fraud Detection Rules
- Section 5.1 — Alert Notifications
Correct? (yes / adjust)
Append an entry to specs/brd/changelog.md (create from state/changelog-template.md if it does not exist).
Changelog format:
## v{N} — {YYYY-MM-DD}
- **{addition|modification|removal}:** {change description}
- **Affected sections:** {section list}
- **Requested by:** user
- **Impact:** {pending — filled in Step 5}
Increment the version number from the last entry. The version is a simple integer (v1, v2, v3...).
Apply the change to the relevant section(s) of specs/brd/brd.md. Mark the updated section with a comment:
<!-- Updated in v{N} — {change description} -->
Trace the change through the project artifacts. Check each of these and report whether it is affected:
| Artifact | Location | How to check |
|---|---|---|
| Epics & stories | specs/features.json | Search for stories referencing affected BRD sections |
| Design docs | specs/design/ | Search for references to changed requirements |
| UI mockups | specs/design/mockups/ | Check if mockup covers affected feature |
| Test plans | specs/tests/ | Search for test cases covering affected behavior |
| Test code | tests/ | Search for test files testing affected stories |
| Implementation | src/ | Search for code implementing affected stories |
| Playwright E2E | tests/e2e/ | Search for E2E tests covering affected flows |
Present the impact summary:
Impact Analysis for v{N}:
Stories affected: 3 (S-012, S-015, S-022)
Design docs: 1 (fraud-rules-arch.md)
Mockups: 1 (alert-dashboard.html)
Test plans: 2
Test files: 4
Source files: 6
E2E tests: 1
Estimated rework: Medium
Update the changelog entry's **Impact:** line with the summary.
Only re-run the phases that are actually affected. Do not re-run the entire pipeline.
Cascade scope rules:
| Change type | What gets re-run |
|---|---|
| New requirement (addition) | spec-writer (new stories), architect --amendment, ui-mockup (if UI), test-engineer (new cases), mark new stories pending in features.json |
| Modified requirement | spec-writer (update stories), architect --amendment (if arch impact), ui-mockup (if UI change), test-engineer (update cases), mark affected stories pending |
| Removed requirement | spec-writer (remove/archive stories), test-engineer (remove cases), mark affected stories removed |
For each phase in the cascade:
All agents check specs/brd/changelog.md at the start of their run. If the changelog version is newer than the version recorded in their last output, they print a warning:
WARNING: BRD has changed (v{N}) since this artifact was last generated (v{M}). Run
/changeto cascade updates or acknowledge the delta.
This convention is documented here so agents can implement it consistently.
Version tracking: Each agent output file includes a frontmatter or header comment with brd-version: v{N}. When the agent runs, it compares this against the latest changelog version.
# BRD Changelog
## v3 — 2026-04-11
- **modification:** Fraud score threshold changed from 0.7 to 0.85
- **Affected sections:** 3.2, 3.4
- **Requested by:** user
- **Impact:** 2 stories, 1 design doc, 3 test files, 4 source files
## v2 — 2026-04-09
- **addition:** Add real-time webhook notifications for high-risk transactions
- **Affected sections:** 5.1
- **Requested by:** user
- **Impact:** 3 new stories, 1 mockup, 2 test plans
## v1 — 2026-04-07
- **Initial BRD approved**
specs/brd/changelog.md does not exist, create it from state/changelog-template.md and set the initial version to v1./change skill does NOT re-run code generation automatically. It marks stories as pending so the next /implement or /build run picks them up./change multiple times, and the cascade will accumulate. Run the cascade once when ready.