Repository navigation
Releases
v0.1.6 — Tested installation and credential guidance
v0.1.6 — actionable missing-credential guidance
Submitting a task without the configured environment API key now shows which variable is missing and how to configure or restart Pan. The task stays failed; no request is sent without credentials. Keychain-specific guidance does not incorrectly ask users to set an environment variable. Unknown provider errors remain generic; raw errors and secrets are never included in the new notice.
Download pan-agent-0.1.6.tgz and SHA256SUMS, then follow the installation README. No compiler, repository checkout, Python or new runtime dependency is required. MIT license; existing Markdown and prior demonstration releases remain available.
This is a narrow presentation fix discovered by independent installation/configuration testing. It does not change authentication, model/tool execution, retry or archive semantics. Evidence.
v0.1.5 — Real-time Markdown in the installed CLI
v0.1.5 — real-time Markdown demonstration
A 29-second original-speed recording shows the normal installed Pan CLI rendering a real Kimi response as it arrives. The run uses ordinary configuration, one text-only prompt and zero tools. No fixture provider, replay or artificial token pacing is used.
Markdown is already built into v0.1.4. Install that unchanged package using README; configure your API credential and start the TUI normally. This release adds video, Chinese subtitles and documentation, not a new runtime package.
Download pan-agent-markdown-live.mp4, pan-agent-markdown-live.srt and SHA256SUMS. Recording and supported scope. Existing offline examples and all prior releases remain available and clearly labeled.
v0.1.4 — Markdown rendering in the TUI
v0.1.4 — readable assistant Markdown in the TUI
Assistant responses and their archived replay now format headings, lists, bold, inline code and fenced code blocks. Partial streaming text remains readable. Tool output, permission prompts and stored run records keep their original semantics and literal text.
Download pan-agent-0.1.4.tgz and SHA256SUMS, verify the checksum and install with npm as described in README. markdown-before.png and markdown-after.png compare actual TUI replay of the same offline fixture. No new paid model or benchmark run was used.
This is a limited presentation feature, not full CommonMark. NO_COLOR and TERM=dumb suppress styles; unsupported Markdown stays readable. No new runtime dependencies. MIT license; macOS arm64 validated, other platforms remain unverified.
v0.1.3 — Original Harbor evaluation demonstration
v0.1.3 — Original Harbor evaluation demonstration
Watch Pan recover an original Terminal-Bench database task, run the original Harbor verifier (7/7 passed, reward 1), clean up the container and replay the same saved run. The 2:31 video includes Chinese explanations, readable detail views and explicit live/archive/speed labels.
Download pan-agent-harbor-demo.mp4, matching UTF-8 pan-agent-harbor-demo.srt and SHA256SUMS. Runtime installation remains the unchanged v0.1.0 package; this is a demonstration/media release.
This is one separately authorized new run after an external launcher repair. The earlier timeout attempt remains unscored and preserved. It is not an overall benchmark result. Actual integration uses Harbor DockerEnvironment + Verifier APIs, with the installed TUI on Mac and tools in an SSH-connected server container. Container-scoped per-request approvals are automated; original 900-second phase limits and the inherited 64-round limit remain in effect.
Recording, provenance and limitations. Terminal-Bench task by Xiangning Lin, Apache-2.0; Pan-authored code MIT.
Pan Agent — 逐幕讲解字幕版
Pan Agent — 逐幕讲解字幕版
在已验收的 SQLite WAL 恢复演示上，为全部11幕添加解释字幕：当前做什么、为什么做，以及结果能说明什么。白色字幕说明动作，蓝色字幕补充背景；介绍 WAL、校验和、字节序、检查点处理和 Trace 回放。保留关键细节放大、两次真实错误，以及 LIVE / ARCHIVE DETAIL / 速度标签。
视频129.8秒、1920×1080；字幕已烧录到视频，SRT提供相同文字和时序。复用v0.1.1同一运行的原始录制，本次没有新增模型调用或任务执行，没有修改Runtime。历史演示和发布资产保持不变。
该任务基于Terminal-Bench 2.0的db-wal-recovery，采用本地路径适配，非新增官方评测成绩。原任务恢复11条记录，适配检查7/7通过；归档回放不重新调用模型或工具。字幕中的事后检查和历史回放均明确标示。
这是媒体与文档版本。安装仍使用v0.1.0 Runtime包，参见README。Pan代码MIT许可，上游任务保留Apache-2.0归属；本次不分发字体二进制或原始私有记录。
Human提出解释性字幕要求，Builder制作字幕与渲染，独立Regulator复核画面、来源及时序。SHA256SUMS列出MP4与SRT的校验值。
Pan Agent — WAL recovery, failure correction and Trace replay
Pan Agent — database recovery demonstration
Watch the 129.8-second TUI demonstration: investigate an unreadable SQLite WAL, correct a real checksum-endianness error, recover all 11 records, and inspect the same run through Trace replay. Key evidence is enlarged and annotated; speed changes and archived detail views are labelled.
This is a media/documentation release. The executable is unchanged: download the v0.1.0 runtime package and follow the README.
One real Kimi Code task produced 11 model exchanges and 11 tool results, including two natural errors. Known token usage: 49,050 input + 3,806 output = 52,856; unknown usage records: 0. Original live capture lasted 371 seconds including interaction and replay; idle/navigation time is shortened in this edit. The independent review verified the recovery scripts against original inputs, 7/7 path-adapted upstream tests, a negative test, and replay with zero provider/tool/network calls.
Based on Terminal-Bench 2.0's db-wal-recovery task, authored by Xiangning Lin. This fresh macOS demonstration adapts the original /app path and is not a new official benchmark score. Both model errors are retained in the original trace. The video comes from actual PTY output, not a fabricated UI. Raw records and test answers are not public assets.
Task provenance, adaptations and edit map · Independent accepted Verdict
Pan-authored code is MIT; the upstream Terminal-Bench task retains its Apache-2.0 license. Human specified the task framing and visual emphasis; Kimi generated the recovery implementation; Builder recorded/edited and Regulator independently checked evidence. No runtime reliability or benchmark improvement is inferred from this demonstration.
Assets: pan-agent-wal-demo.mp4 and SHA256SUMS. Previous v0.1.0 assets remain unchanged.
Pan Agent v0.1.0 — TUI learning release
Pan Agent v0.1.0 — TUI learning release
Release tag: v0.1.0. Download the assets from the GitHub release linked below.
Pan Agent is a local terminal runtime with file/shell tools, per-operation approval and sealed run archives. This release includes an offline scripted walkthrough and supports real tasks through configured DeepSeek or Kimi Code credentials.
Assets
pan-agent-0.1.0.tgz: prebuilt JavaScript TUI package. Node >=22.19.0 and npm required; no compiler, Docker, database, Go or Python runtime requirement.pan-agent-demo.mp4: accepted 41-second demonstration of one real Kimi file task followed by a separately captured replay of the same run. Input automated, idle waits shortened; no fabricated UI.SHA256SUMS: SHA-256 checksums of the package and video.
Release: https://github.com/pym96/Pan-agent/releases/tag/v0.1.0
Quick start
Download the tarball and checksum file from the release, verify the package's SHA-256, then:
npm install ./pan-agent-0.1.0.tgz --omit=dev --ignore-scripts --no-audit --no-fund
./node_modules/.bin/pan-agent demoThe demo uses scripted model responses and real local tools; no API key or provider call. For real tasks, run pan-agent configure, configure a credential source, and launch with explicit workspace/archive directories as documented in README.
Boundaries
Validated on macOS arm64, Node v26.7.0 and an interactive PTY. Node >=22.19.0 is the declared engine range; the exact minimum, Linux and Windows have not been verified for this release. Tools use host-user authority: workspace is a working directory, not a sandbox. Replay is read-only inspection, not process/session resume. No production-platform maturity, benchmark score or reliability improvement is claimed.
License and attribution
Pan-authored code: MIT, copyright 2026 Pan Agent contributors. The package includes LICENSE; dependencies and external material keep their respective licenses. Human set architecture/scope and licensing direction; Coding Agents implemented and independently reviewed the candidate. The licensed package preserves the accepted Stage A executable bytes and video.