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。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/majiayu000/claude-skill-registry --skill trace命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
LLM token logprobs and calibration. Per-decision confidence, ECE, Brier, reliability diagrams, low-confidence triage.
Analyze LLM token logprobs and calibration. Use for per-decision confidence, ECE, Brier scores, reliability diagrams, and low-confidence triage.
回顾最近 N 天的 Claude Code 使用记录——扫描原始会话数据,按主题分组汇总"我都做了什么",并从个人操作系统视角输出模式、风险与增删建议。当用户说 /recap、"看看我这几天做了什么"、"回顾一下我最近的会话"、"这两天我用 claude 干了啥"、"活动回顾" 时使用。
基于 SOC 职业分类
正在显示 SKILL.md
| name | trace |
| description | Trace decisions through artifacts. |
| practices | ["distributed-tracing","adr","ebpf-observability"] |
| hexagonal_role | supporting |
| consumes | [] |
| produces | ["result.json"] |
| context_rel | [{"kind":"customer-of","with":"provenance"}] |
| skill_api_version | 1 |
| allowed-tools | Read, Grep, Glob, Bash |
| context | {"window":"fork","intent":{"mode":"task"},"sections":{"exclude":["HISTORY"]},"intel_scope":"full"} |
| metadata | {"tier":"knowledge","dependencies":["provenance"]} |
| output_contract | stdout: decision provenance chain |
Quick Ref: Trace design decisions through CASS sessions, handoffs, git, and artifacts. Output:
.agents/research/YYYY-MM-DD-trace-*.md
YOU MUST EXECUTE THIS WORKFLOW. Do not just describe it.
For knowledge artifact lineage (learnings, patterns, tiers), use /provenance instead.
CLI dependencies: cass (session search). If cass is unavailable, skip transcript search and rely on git log, handoff docs, and .agents/ artifacts for decision tracing.
Given /trace <concept>:
Determine what kind of provenance to trace:
IF target is a file path (contains "/" or "."):
→ Use /provenance (artifact lineage)
IF target is a git ref (sha, branch, tag):
→ Use git-based tracing (Step 2b)
ELSE (keyword/concept):
→ Use design decision tracing (Step 2a)
Launch 4 parallel search agents (CASS, Handoff, Git, Research) and wait for all to complete.
Backend: Agents use Task(subagent_type="Explore") which maps to task(subagent_type="explore") in OpenCode. See skills/shared/SKILL.md ("Runtime-Native Spawn Backend Selection") for the shared contract.
Read references/discovery-patterns.md for agent definitions and prompts.
Read references/discovery-patterns.md for git-based tracing commands.
Merge results from all sources into a single chronological timeline (oldest first). Deduplicate same-day/same-session events. Every claim needs a source citation.
For each event in timeline, identify:
Write to: .agents/research/YYYY-MM-DD-trace-<concept-slug>.md
Read references/report-template.md for the full report format and deduplication rules.
Tell the user:
Read references/edge-cases.md for handling: no CASS results, no handoffs, ambiguous concepts (>20 results), and all-sources-empty scenarios. General principle: continue with remaining sources and note gaps in the report.
.agents/research/ artifact| Skill | Purpose | Input | Output |
|---|---|---|---|
/provenance | Artifact lineage | File path | Tier/promotion history |
/trace | Design decisions | Concept/keyword | Timeline of evolution |
Use /provenance for: "Where did this learning come from?"
Use /trace for: "How did we decide on this architecture?"
# Trace a design decision
/trace "three-level architecture"
# Trace a role/concept
/trace "Chiron"
# Trace a pattern
/trace "brownian ratchet"
# Trace a feature
/trace "parallel wave execution"
User says: /trace "agent team protocol"
What happens:
.agents/research/2026-02-13-trace-agent-team-protocol.md with full timeline and citationsResult: Complete evolution timeline showing how agent team protocol developed across 7 sessions with source citations.
User says: /trace abc1234
What happens:
git log --grep to find related work.agents/ for contemporary research/plansResult: Trace report links commit to broader design context from surrounding artifacts.
| Problem | Cause | Solution |
|---|---|---|
| CASS returns no results | Session search not installed or query too specific | Check which cass. If missing, skip CASS and rely on handoffs/git/research. Try broader query terms. |
| Timeline has gaps | Not all decisions documented in searchable artifacts | Note gaps in report. Suggest interviewing team members or checking Slack/email archives for missing context. |
| Too many results (>50 matches) | Very broad concept or high-frequency term | Read references/edge-cases.md for ambiguous concept handling. Narrow query or filter by date range. Ask user for more specific aspect to trace. |
| Empty trace report (all sources failed) | Concept genuinely undocumented or typo | Verify spelling. Try synonyms. Report to user: "No documented history found. This may be a new concept or may need different search terms." |