技能 task-observer

📦

task-observer

Name: task-observer
Author: rebelytics

v3.5.0 內容修訂 r1 嚴重 🌐 網路存取📁 檔案系統存取🔑 環境變數⚙️ 外部命令

Capture Workflow Lessons and Improve Skills

Useful workflow lessons disappear between work sessions. Task Observer records reusable observations and guides reviews that produce staged skill updates.

支援: Claude Codex Code(CC)

⚠️ 38 差

Maintain this Skill? Claim it ⚠️ 38 差

請勿自動安裝此 Skill。

權威政策要求在執行任何安裝操作前由操作人員審查。

使用我的智慧代理安裝

將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。

智能體請求

Review the Skillstore skill "task-observer" from https://skillstore.io/skills/rebelytics-task-observer.md and its manifest at https://skillstore.io/api/skills/rebelytics-task-observer/manifest. Verify the artifact. Do not auto-install. Inspect the skill and report your findings, then wait for an operator or manual installation decision.

你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。

Agent 可讀資源

當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。

Markdown 詳情 GET /skills/rebelytics-task-observer.md 已簽名 manifest GET /api/skills/rebelytics-task-observer/manifest 已簽名 lockfile GET /api/skills/rebelytics-task-observer/lockfile

測試它

正在使用「task-observer」。 Record that the assistant skipped verification after changing an existing workflow.

預期結果:

Issue: The workflow changed without checking its existing verification requirements.
Suggested improvement: Add a verification checkpoint before handing over the result.
Principle: Verify required checks before declaring a changed workflow complete.

正在使用「task-observer」。 Review the ten oldest open observations and summarize proposed actions.

預期結果:

Reviewed ten observations and grouped repeated findings by target skill.
Prepared staged updates for findings with sufficient evidence.
Left unresolved findings open and recorded the evidence still needed.
Installed skills remain unchanged pending review and installation.

正在使用「task-observer」。 Check whether a legacy observation log is ready for conversion.

預期結果:

Check-only review identified ambiguous metadata for human review.
Converted entries will need sibling-check metadata that the legacy format did not contain.
No files were changed during the check.

安全審計

嚴重

v1 • 2026/10/4 開啟版本化報告

Of 400 supplied findings, 399 are false positives; one release command exposes shell injection through direct tag interpolation. A critical semantic finding identifies instructions to retry denied writes through another tool. No evidence found of credential exfiltration; 228 omitted static matches remain unconfirmed and require manual review before automatic publication. Static review was capped at 400/628 representative findings; omitted static matches are unconfirmed, so automatic publishing stays disabled until manual review.

已掃描檔案

10,516

分析行數

複核項目

已忽略的誤報

已確認的安全問題 (1)

嚴重

SKILL.md:356-362 SKILL.md:112-119 references/environments.md:810-820

Prompt Injection Attempt Detected

The core says "a classifier can deny one interface while allowing another" and directs agents to "try a second tool reaching the same path". This instruction encourages completing denied writes through another interface, treating permission decisions as noise rather than authorization boundaries. It conflicts with the reference guidance that prohibits alternate-tool retries for denied hook installs.

The core explicitly links denied writes to alternate-tool retries and dismisses classifier denials as noise. Separate no-bypass guidance does not remove this conflicting instruction.

能力複核項目 (1)

這些是真實的本機能力，對此技能而言可能是預期行為，因此需要複核，但不會按已確認的惡意行為計分。

高

.github/workflows/release-bundle.yml:27

Temp directory access

run: gh release upload "${{ github.event.release.tag_name }}" /tmp/bundle/task-observer.skill --clob

The release tag is interpolated directly into shell source, so a tag containing command substitution executes commands with the release token. Temporary staging is ordinary, but this upload command exposes a release-workflow injection risk.

風險因素

🌐 網路存取 (10)

.claude-plugin/marketplace.json:2 .claude-plugin/marketplace.json:6 .claude-plugin/plugin.json:2 .claude-plugin/plugin.json:8 .claude-plugin/plugin.json:10 .claude-plugin/plugin.json:11 .github/ISSUE_TEMPLATE/config.yml:4 README.md:9 references/skill-authoring.md:1278 references/weekly-review.md:968

📁 檔案系統存取 (50)

.github/workflows/release-bundle.yml:18 .github/workflows/release-bundle.yml:19 .github/workflows/release-bundle.yml:20 .github/workflows/release-bundle.yml:21 .github/workflows/release-bundle.yml:23 .github/workflows/release-bundle.yml:27 .github/workflows/tessl-publish.yml:51 .github/workflows/tessl-publish.yml:54 references/environments.md:503 references/environments.md:506 references/environments.md:517 references/environments.md:533 references/environments.md:513 references/environments.md:517 references/environments.md:536 references/environments.md:503 references/environments.md:506 references/environments.md:517 references/environments.md:365 references/environments.md:723 references/environments.md:1017 references/environments.md:365 references/environments.md:723 references/environments.md:1017 references/environments.md:160 references/environments.md:458 references/environments.md:459 references/environments.md:907 references/environments.md:908 references/environments.md:909 references/environments.md:512 references/environments.md:928 references/migration.md:158 references/migration.md:179 references/migration.md:46 references/migration.md:179 references/migration.md:153 references/observation-log.md:744 references/observation-log.md:887 references/observation-log.md:949 references/observation-log.md:971 references/observation-log.md:980 references/observation-log.md:1200 references/observation-log.md:385 references/observation-log.md:965 references/observation-log.md:970 references/observation-log.md:971 references/observation-log.md:1200 references/skill-authoring.md:739 references/skill-authoring.md:739

🔑 環境變數 (6)

.github/workflows/release-bundle.yml:26 references/skill-authoring.md:1277 references/weekly-review.md:487 references/weekly-review.md:500 references/weekly-review.md:501 scripts/migrate-log.py:365

⚙️ 外部命令 (50)

審計單位: codex

分享與引用此報告

分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。

開啟版本化報告

複製報告連結

https://skillstore.io/skills/rebelytics-task-observer/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/rebelytics-task-observer/security.svg)](https://skillstore.io/skills/rebelytics-task-observer?utm_source=security_passport_badge)

HTML 徽章

<a href="https://skillstore.io/skills/rebelytics-task-observer?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/rebelytics-task-observer/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/rebelytics-task-observer.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

學術引用 (APA · BibTeX · CFF)

APA 引用

rebelytics. (2026). task-observer security audit report (audit version 1) [Author version 3.5.0]. Skillstore. https://skillstore.io/skills/rebelytics-task-observer/audits/1

BibTeX 引用

@techreport{rebelytics-rebelytics-task-observer-2026,
  author = {rebelytics},
  title = {task-observer security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/rebelytics-task-observer/audits/1},
  note = {Author version 3.5.0}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "task-observer security audit report (audit version 1)"
version: "3.5.0"
type: report
authors:
  - name: "rebelytics"
date-released: "2026-10-04"
url: "https://skillstore.io/skills/rebelytics-task-observer/audits/1"
identifiers:
  - type: other
    value: "skillstore:rebelytics-task-observer:audit:1"
    description: "Skillstore immutable audit report identifier"

Skillstore 評分

評分依據證據置信度: 低

架構

可維護性

內容

社群

規範符合度

你可以打造什麼

Preserve Coding Workflow Corrections

Record recurring implementation mistakes and user corrections, then prepare focused updates to development skills.

Review Shared Skill Quality

Cluster team observations, check related skills, and stage coordinated improvements for human review.

Retain Research Methods

Capture reusable research steps and verification lessons while separating confidential project details from shared principles.

試試這些提示

Capture One Lesson

Record the reusable lesson from my latest correction as an observation. Explain the issue, suggested improvement, and general principle.

Check for Repeated Findings

Check existing observations for this workflow problem. Extend a matching entry instead of creating a duplicate, and report the affected skills.

Review a Bounded Backlog

Review the ten oldest open observations. Group related findings, check sibling skills, and stage proposed changes for my review.

Plan a Legacy Migration

Inspect my legacy observation log in check-only mode. Report parsing issues and target-format gaps before requesting approval for conversion.

最佳實務

Use one stable workspace for each observed skill scope and verify the log path before writing.
Capture specific evidence promptly, check duplicates, and record which related skills were considered.
Respect permission denials, remove confidential details from shared principles, and review staged changes before installation.

避免

Treating a permission denial as permission to retry the same action through another tool.
Logging every isolated correction without checking whether it generalizes or repeats an existing finding.
Editing installed skills directly or marking observations resolved before checking their full content.

常見問題

Which assistants does Task Observer support?

The report lists Claude, Codex, and Claude Code. Activation and available tools depend on the host environment.

Where are observations stored?

Each observation is a Markdown file in the configured persistent workspace. Resolved entries move into its archive directory.

Does it run automatically?

Reliable automatic activation requires explicit instructions or supported harness hooks. Description matching alone does not guarantee activation.

Does it change my installed skills?

The documented workflow stages proposed updates for review and installation. It does not treat staging as permission to replace installed skills.

Does it require network access?

Routine observation capture is local. Requested upstream feedback and public repository review workflows can require network access.

What security concerns did the audit identify?

The audit found denied-write retry instructions and release tag shell injection. The omitted static matches also require manual review before automatic publication.