菜单
Run the 5-stage Token Transformation Pipeline (TRAE §5.2.2) over the current session state — Collection, Ranking, Compression, Budgeting, Assembly. Emits a concrete next-action recommendation (/compact, /lean-md, or /progress-log + fresh session) instead of a generic warning.
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
| name | token-pipeline |
| description | Run the 5-stage Token Transformation Pipeline (TRAE §5.2.2) over the current session state — Collection, Ranking, Compression, Budgeting, Assembly. Emits a concrete next-action recommendation (/compact, /lean-md, or /progress-log + fresh session) instead of a generic warning. |
| when_to_use | Fires automatically via track-context-pressure at ~65% pressure; invoke manually any time context feels heavy or auto-compact looks imminent. Do NOT use for a holistic context audit before work starts — use `/audit-context` instead; this skill is the in-flight pressure-relief decision. |
| disable-model-invocation | true |
| allowed-tools | ["Read","Glob","Grep","Bash"] |
| logical | next-action recommendation emitted (/compact, /context-tricks, or continue) with quantified pressure value |
Replace the "context is heavy" feeling with a structured report. TRAE §5.2.2 defines the pipeline:
Inventory the load. For each source, estimate size in lines / rough tokens:
| Source | Check |
|---|---|
| CLAUDE.md (root + home) | wc -l CLAUDE.md ~/.claude/CLAUDE.md |
| MCP instructions | Check mcp-instruction-monitor.sh log (context-engine) for current MCP context reminders |
| Memory index | wc -l .claude/memory/MEMORY.md if present |
| Plan + spec + features | wc -l .claude/plans/*.md .claude/spec.md and jq length .claude/features.json |
| Progress tail | wc -l claude-progress.txt |
| System reminders | Check track-system-reminders.sh log |
Print a table with line counts.
For each source, score on two axes (1–5):
Flag low-score + high-size entries as "Lost in the Middle" candidates.
For each source with score < 6 (low recency + low relevance):
/lean-md/progress-log + /compact/memory-index (memory plugin)Suggest a per-category cap (tokens):
| Category | Suggested cap |
|---|---|
| CLAUDE.md total | 3000 |
| Active plan | 2000 |
| Active spec | 2000 |
| Memory pointers | 500 |
| Progress tail (3 entries) | 1500 |
| Current turn work | remainder |
Emit ONE concrete recommendation, chosen by highest expected payoff:
/lean-md/progress-log + /compact/memory-index audit/compact with preservation instructions/progress-logPrint the recommendation in a boxed line so it's hard to miss.
/lean-md, /progress-log + /compact, plan trim, /memory-index)plugins/context-engine/hooks/track-context-pressure.sh at P2 / S2 thresholds (≈65% pressure / ~15 exchanges)..claude/plans/, .claude/spec.md, .claude/features.json, claude-progress.txt./compact, /lean-md, /progress-log, /memory-index./rest-audit Efficiency axis consumes this skill's recommendations./lean-md if CLAUDE.md is the dominant load.jq unavailable → skip features.json stage; note in output./compact or /lean-md. The user should choose; this skill is advisory.Use when the user describes a multi-step feature or refactor and you need to decide whether to handle it solo, dispatch a parallel `/fan-out`, or run a `/worktree-team` planner→generator→reviewer pipeline. Outputs a routing recommendation with the reasoning behind it.
Use whenever the user asks to apply the same operation to many independent files or components — bulk migration, batch refactor across controllers, parallel exploration of unrelated subsystems. Dispatches one subagent per file with a shared prompt template, then collects the results.
Use to toggle the safe-mode flag at .claude/safe-mode. The flag is written automatically by consecutive-failure-guard once tool failures hit FORGE_SAFE_MODE_THRESHOLD (default 5), and block-destructive.sh denies every Bash/Write/Edit while it exists. /safe-mode off clears the flag, logs the exit, and prompts /postmortem; /safe-mode on enters manually; /safe-mode status reports current state.
Audit or scaffold a CLAUDE.md against the Karpathy 4-section structure (Think Before Coding · Simplicity First · Surgical Changes · Goal-Driven Execution). Companion to /lean-md — structure audit, not size trim.
Print every policy enforcement point declared in plugins/diagnostics/registry/policies.json — id, verdict, plugin, hook event, severity, bypass — grouped by verdict. Single discoverable index of what the harness blocks, anchors, nudges, or logs at runtime.
Audit the project against the R.E.S.T. framework (TRAE) — Reliability, Efficiency, Security, Traceability. Reads ledger entries, hook state, and artifact presence across long-session + policy-gateway + context-engine + traces. Single PASS/WARN/FAIL table. Cross-cut meta-surface.