Viewtrace AI — AI skill for Claude Code
See how your coding agent actually worked.
How to install Viewtrace AI
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open hyukvoid/viewtrace-ai and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Viewtrace AI does
See how your coding agent actually worked. A tiny local CLI that turns Claude Code / Codex sessions into flight reports: edits, verification runs, FAIL->PASS loops, and session comparisons. Local, read-only, no API key, nothing uploaded.
Alternatives in AI
- Omniget — Udemy & Hotmart course downloader, YouTube downloader (yt-dlp GUI, 1,800+ sites) + desktop app for AI agents 13.9k ★
- UltraCode Shim — Give Claude Code's ultracode mode to ANY model you already pay for 422 ★
- Prompt Tower — Context management for long-context LLMs, agents, and vibe coding 382 ★
README
Agent Pigeon
**ViewTrace AI is under construction on top of this repository.** The new `viewtrace` bin traces the evidence behind AI answers for research/comparison/recommendation runs. Status is honest and early: batch ingest **and live collection** of the reference JSONL format are implemented and tested — `viewtrace up` starts the local collector, `viewtrace run -- ` wraps a reference producer and shows activity live, `status`/`runs`/`replay` report the honest state. Final public receipts, safe answer resolution, a recent picker and the answer-first loopback report are implemented. The shared evidence analyzer supplies seven mode lenses, claim/evidence relationships, source provenance and unresolved areas for CLI and Web inspection. Real agent adapters are **not** supported yet (`viewtrace adapters` shows the honest capability matrix). Requires Node >= 22.13 (built-in `node:sqlite`).
- Local-only. Everything ViewTrace records lives under one data root (default
~/.viewtrace); the runtime makes zero external network requests — no cloud, accounts, telemetry or API keys. The collector's control channel binds127.0.0.1only and requires a per-process token.- Original agent history files are read-only inputs and are never modified. ViewTrace's own recordings are local writes under the data root (SQLite
viewtrace.db, per-runtrace.jsonland a sanitized live spool, mode 0700/0600 on POSIX). Explicit delete/keep/prune commands control individual runs; default retention is indefinite. Delete the data root to erase everything ViewTrace recorded.- Private reasoning is never collected. Declared thinking/analysis payloads are stripped before anything touches disk; provenance labels (agent-reported / observed / inferred) are preserved exactly as claimed, never promoted.
- The legacy
agent-pigeonflight recorder below is unchanged.
ViewTrace: Answer → Curiosit
Related Skills
Headsign
A tiny phase gate for coding agents: your shell checks decide pass/fail (never the LLM), your workflow routes
Warden Receipt
Show token-warden rule receipts — the per-rule verdict card with token savings vs. rent, per-task pass/fail, t
Tausik Core
AI coding agents can't quietly fake "done". Signed ed25519 receipts prove the gates actually ran — bound to th
Tracon
The flight recorder for AI coding agents. See every command, file edit, and package install Claude Code, Codex
Token Ledger
Know what your AI coding agents actually cost. A tiny macOS menu bar app for Claude Code, Codex, Droid, Pi, De
Claude Image
教 Claude Code/Codex 真正用好 gpt-image-2 的 skill 包 · Drop-in skill teaching agents to actually use GPT Image 2 — i
Related Agents
Repo Orientation
Read-only orientation for a codebase you have not worked in. Traces how the app actually starts and where its
Omarchy Coverage Reporter
Use this agent when an Omarchy plugin verification reports PASS and you need to establish what actually ran, w
V Pre Flight Runner
Runs /v pre-flight quality gates (TSC, PHPStan, lint, build, pest, vitest, audits) and writes PRE_FLIGHT_REPOR