技能 gaffa-find

📦

gaffa-find

Name: gaffa-find
Author: gaffaai

內容修訂 r1 高風險 ⚙️ 外部命令🌐 網路存取📁 檔案系統存取🔑 環境變數

Find Website Information With Gaffa

Finding reliable website information often requires repeated page inspection and extraction attempts. This skill guides a bounded Gaffa workflow that validates results and returns evidence with a reusable script.

支援: Claude Codex Code(CC)

⚠️ 38 差

Maintain this Skill? Claim it ⚠️ 38 差

安裝前請停止並請求確認。

請審查安裝計畫，並在變更檔案前取得使用者明確同意。

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 "gaffa-find" from https://skillstore.io/skills/gaffaai-gaffa-find.md and its manifest at https://skillstore.io/api/skills/gaffaai-gaffa-find/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

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/gaffaai-gaffa-find.md 已簽名 manifest GET /api/skills/gaffaai-gaffa-find/manifest 已簽名 lockfile GET /api/skills/gaffaai-gaffa-find/lockfile

測試它

正在使用「gaffa-find」。 Find the chief executive on a public company leadership page.

預期結果:

Illustrative result: success.
Candidate: Morgan Lee, listed as Chief Executive Officer.
Evidence: the leadership page identifies the name and role in the main content region.
Deliverable: a reusable Gaffa extraction script with request recording disabled.

正在使用「gaffa-find」。 Find a published salary range when the available pages do not provide clear evidence.

預期結果:

Illustrative result: needs-human-review.
No verified salary range was found within the request budget.
The final response identifies the partial candidate, log location, and failed request identifier when available.
The answer does not invent a salary range.

安全審計

高風險

v1 • 2026/10/14 開啟版本化報告

Most alerts concern Markdown formatting or intended API operations; confirmed risks include conditional redaction, unsafe sed interpolation, exposed credential arguments, and predictable temporary files. The logging example persists content before redaction, and mandatory browser recording can retain sensitive pages externally. No evidence of prompt injection or intentional exfiltration was found.

已掃描檔案

283

分析行數

複核項目

已忽略的誤報

已確認的安全問題 (2)

高

templates/loop.md:102-113 SKILL.md:80-86

Logging Template Persists Data Before Redaction

The template writes log content before scrubbing it, reversing the explicit redaction-first rule. When adapted to real iteration data, interruptions or scrub failures can leave sensitive content on disk.

The shell example writes at line 102 before running sed at line 110, contradicting SKILL.md lines 85-86. Its placeholder requires adaptation, but the demonstrated write order is directly observable.

中

SKILL.md:40-43 SKILL.md:157-159

Mandatory Recording Can Retain Sensitive Page Content

The skill requires recording every browser request and states retention lasts seven days to three months. Sensitive pages can be retained externally without a documented sensitivity check or explicit recording consent.

Recording and retention are explicit requirements, and neither file provides a sensitivity check or recording consent step. Exposure depends on the target content; no malicious intent is established.

能力複核項目 (7)

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

SKILL.md:82

Generic API/secret keys

4. If the runtime value of `GAFFA_API_KEY` appears as a substring anywhere in a payload you are abou

The substring scrub is conditional on length and character composition in lines 83-84. Other nonempty keys can remain in unstructured payloads despite field-name redaction.

高

templates/loop.md:10

Generic API/secret keys

-H "X-API-Key: ${GAFFA_API_KEY}" \

The shell expands GAFFA_API_KEY into the curl authentication argument, despite using the legitimate vendor endpoint. Local process inspection can expose credentials where argument access is permitted.

高

templates/loop.md:25

Generic API/secret keys

-H "X-API-Key: ${GAFFA_API_KEY}" \

The expanded API key becomes part of the curl command-line header argument. It can be exposed to local process inspection, although the network destination is legitimate.

高

templates/loop.md:53

Generic API/secret keys

-H "X-API-Key: ${GAFFA_API_KEY}" \

Environment expansion places the API key directly in the curl process arguments. Systems allowing other users or monitoring tools to inspect arguments can expose the credential.

高

templates/loop.md:108

Generic API/secret keys

if [ "${#GAFFA_API_KEY}" -ge 16 ] && printf '%s' "$GAFFA_API_KEY" | grep -q '[0-9]' && printf '%s' "

The conditional skips substring redaction for short, digit-only, or letter-only keys. Those credentials can survive in unstructured logs before the unconditional rename.

高

templates/loop.md:110

Generic API/secret keys

sed -i "s|${GAFFA_API_KEY}|<REDACTED>|g" "$TMPLOG"

The key enters a sed regex and process argument without literal escaping or protection from argument inspection. Scrubbing can fail or miss matches, yet line 113 still renames the log.

中

templates/loop.md:99

Hidden file access

FINALLOG="./.gaffa-find-$(date +%Y%m%dT%H%M%S).log"

Lines 99-102 derive a predictable temporary filename and open it with ordinary redirection. A precreated symlink can redirect writes in an attacker-writable directory.

風險因素

⚙️ 外部命令 (50)

🌐 網路存取 (12)

SKILL.md:33 SKILL.md:45 SKILL.md:47 SKILL.md:93 SKILL.md:97 SKILL.md:149 templates/loop.md:9 templates/loop.md:13 templates/loop.md:24 templates/loop.md:29 templates/loop.md:52 templates/loop.md:57

📁 檔案系統存取 (8)

SKILL.md:85 SKILL.md:153 SKILL.md:85 SKILL.md:153 templates/loop.md:90 templates/loop.md:99 templates/loop.md:95 templates/loop.md:96

🔑 環境變數 (11)

SKILL.md:28 SKILL.md:75 SKILL.md:77 SKILL.md:78 SKILL.md:80 SKILL.md:82 templates/loop.md:10 templates/loop.md:25 templates/loop.md:53 templates/loop.md:108 templates/loop.md:110

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/gaffaai-gaffa-find/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/gaffaai-gaffa-find/security.svg)](https://skillstore.io/skills/gaffaai-gaffa-find?utm_source=security_passport_badge)

HTML 徽章

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

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/gaffaai-gaffa-find.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

gaffaai. (2026). gaffa-find security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/gaffaai-gaffa-find/audits/1

BibTeX 引用

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

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "gaffa-find security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
  - name: "gaffaai"
date-released: "2026-10-14"
url: "https://skillstore.io/skills/gaffaai-gaffa-find/audits/1"
identifiers:
  - type: other
    value: "skillstore:gaffaai-gaffa-find:audit:1"
    description: "Skillstore immutable audit report identifier"

Skillstore 評分

評分依據證據置信度: 中

架構

可維護性

內容

社群

規範符合度

你可以打造什麼

Build a Targeted Extraction Script

Turn a website URL and desired field into a validated extraction approach with evidence and a reusable script.

Verify a Public Website Fact

Locate a company fact on public pages and check that the extracted value matches the research question.

Validate Website Data Fields

Extract a structured field and apply an explicit validator before accepting it for a data workflow.

試試這些提示

Find One Public Fact

Use gaffa-find on [website URL] to find [specific fact]. Return the answer, source evidence, and a reusable script.

Extract and Validate a Field

Find [field] on [website URL]. Prefer deterministic extraction and validate the value against [regex or expected type].

Inspect Before Extracting

Use reconnaissance-only mode on [website URL] for [goal]. Return the site map or landing capture without entering the extraction loop.

Refine a Bounded Extraction

Find [information] on [website URL] using [language]. Apply [validator], use at most [iteration count] iterations, and stop for human review if uncertain. Return evidence and a reusable script.

最佳實務

Use authorized public targets, a precise goal, and an explicit validator when correctness matters.
Review request and credit estimates, including proxy bandwidth and recording charges, before running extraction.
Fix credential handling and log redaction, and review recording consent and retention before processing sensitive pages.

避免

Accepting empty or ambiguous extraction output as a verified answer.
Using full-page LLM extraction when a narrow deterministic selector is sufficient.
Copying credential and logging examples unchanged or recording sensitive pages without reviewing retention.

常見問題

What inputs are required?

Provide a target URL and a plain-language information goal. The workflow asks for missing inputs before starting.

Which assistants are supported?

The marketplace lists Claude, Codex, and Claude Code. Execution still requires suitable tools, network access, and Gaffa credentials.

Does it require a Gaffa account?

Yes. The workflow reads GAFFA_API_KEY from the environment and uses paid Gaffa API operations.

Does every extraction use an LLM?

No. It prefers deterministic extraction for stable structures and uses parse_json for interpretive or variable content.

How does the workflow limit spending?

It specifies defaults of three requests per iteration, fifteen requests per invocation, fifty credits, and five iterations. These are agent instructions, not executable enforcement.

What happens when the answer cannot be verified?

It returns needs-human-review or terminal-failure with available evidence and debugging details. It should not invent an answer.