菜单
Work through the legal access hierarchy for paywalled or restricted sources before marking them inaccessible and downgrading confidence
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
Use when an investigation step may need an external integration such as browser acquisition, Maigret account discovery, Junkipedia narrative tracking, OSINT Navigator tool discovery, Noosphere C2PA signing, or Unpaywall access lookup.
Turn investigation findings into explicit, user-approved monitoring recommendations. Spotlight recommends; Mycroft owns any Scoutpost action.
OSINT investigation orchestrator — guides verified investigations from lead to findings to knowledge ingestion. Triggers on "investigate", "investigation", "OSINT", "look into", "dig into".
OSINT investigation toolkit — local SQL index of 12,500 tools (`osint-tools find`, offline), methodology guides, and optional OSINT Navigator (subscription tier). Works offline with any LLM.
Portable claim-to-evidence grounding for knowledge work, fact-checking, and Spotlight investigations. Use when extracting claims or findings, assigning confidence, diagnosing weak support, or deciding whether material is a lead, partially grounded claim, verified finding, disputed claim, or false claim.
Portable safe command construction for any agent skill or recipe. Use before a shell call containing user, model, scraped, configuration, filesystem, or generated values.
| name | content-access |
| description | Work through the legal access hierarchy for paywalled or restricted sources before marking them inaccessible and downgrading confidence |
| version | 1.0 |
| invocable_by | ["investigator","fact-checker"] |
| requires | ["web-archiving","shell-safety"] |
| env_vars | ["CORE_API_KEY","OPENALEX_API_KEY"] |
| attribution | Adapted from jamditis/claude-skills-journalism (https://github.com/jamditis/claude-skills-journalism). Original author: Joe Amditis. MIT License. |
When a source is behind a paywall or access barrier, work through this hierarchy before marking it inaccessible. A paywalled source should not default to low confidence — it should default to an access attempt.
Try each step in order. Stop as soon as you have the full text.
Before any execute-shell command that uses a DOI, URL, query, timestamp, filename, or path, invoke shell-safety and validate the value with scripts/spotlight_safe.py. Prefer --data-urlencode or serialized JSON over interpolating values into shell strings.
External-access boundary: Unpaywall, CORE, Semantic Scholar, web archives, and similar services may receive only the minimum DOI, title/query, URL, or timestamp needed to retrieve the source. Do not send case notes, raw source dumps, unpublished allegations, private source material, credentials, vault contents, or unnecessary personal data. In sensitive mode, remote access services are disabled unless the user explicitly approves an override.
Many paywalled articles have freely distributed copies. Search for them:
search: query="{exact title}" filetype:pdf
search: query="{exact title}" site:{author-institution}.edu
search: query="{exact title}" site:researchgate.net
search: query="{exact title}" site:academia.edu
Unpaywall is optional. Before using it, confirm the integration is green:
execute-shell: python3 integrations/preflight.py --json
Use Unpaywall only when unpaywall is green. It requires $UNPAYWALL_EMAIL in .env; the email is a fair-use identifier, not a notification target. If the integration is red or missing, skip to step 3.
execute-shell: python3 scripts/spotlight_safe.py validate-doi "{DOI}"
execute-shell: curl --get "https://api.unpaywall.org/v2/{DOI}" --data-urlencode "email=$UNPAYWALL_EMAIL"
Parse the response for best_oa_location.url — if present, it's the legal open-access copy. Also check oa_locations[] for mirrors.
If a URL is found, fetch the full text:
fetch: url={oa_location_url}, output_path={CASE_DIR}/research/{filename}.md
execute-shell: curl --get "https://api.core.ac.uk/v3/search/works" \
--data-urlencode "q={query}" \
--data-urlencode "limit=5" \
-H "Authorization: Bearer ${CORE_API_KEY}"
Check results[].downloadUrl for PDF links.
Unkeyed access now sits behind a low shared rate limit and is unreliable for batch lookups. For more than ad-hoc calls, request a free key at semanticscholar.org/product/api and send it as x-api-key; on 429, move to step 5 rather than retrying.
execute-shell: curl --get "https://api.semanticscholar.org/graph/v1/paper/search" \
--data-urlencode "query={query}" \
--data-urlencode "fields=title,openAccessPdf,externalIds" \
--data-urlencode "limit=5"
Check openAccessPdf.url.
The de-facto open scholarly backbone since Microsoft Academic Graph was retired. API-key-required since 2026-02-13 (credit-based rate model — verify at docs.openalex.org). Skip if $OPENALEX_API_KEY is unset.
execute-shell: curl --get "https://api.openalex.org/works"
--data-urlencode "search={query}"
--data-urlencode "per-page=5"
-H "Authorization: Bearer ${OPENALEX_API_KEY}"
Check results[].open_access.oa_url and oa_status (gold/green/hybrid/bronze/closed).
Check if an archived version is accessible before the paywall went up:
execute-shell: python3 scripts/spotlight_safe.py validate-url "{URL}"
execute-shell: curl --get "http://web.archive.org/cdx/search/cdx" \
--data-urlencode "url={URL}" \
--data-urlencode "output=json" \
--data-urlencode "limit=3" \
--data-urlencode "fl=timestamp,statuscode"
execute-shell: python3 scripts/spotlight_safe.py validate-url "{URL}"
execute-shell: curl -sI "https://archive.ph/newest/{URL}" | grep -i '^location:'
A Location pointing at a snapshot (no `/newest/` in it) is the hit; no Location or a `/newest/` Location means no snapshot or CAPTCHA rate-limiting — do not retry in a loop.
Early snapshots often predate paywall implementation. Archive.today caveat (2026): the FBI subpoenaed its registrar in Oct 2025 and Wikipedia stopped accepting it as a citation source in Feb 2026. Prefer Wayback snapshots for anything cited in findings.json; use archive.ph only as fallback and record that in access_notes. If a pre-paywall snapshot is found, retrieve it:
execute-shell: python3 scripts/spotlight_safe.py validate-timestamp "{TIMESTAMP}"
execute-shell: python3 scripts/spotlight_safe.py validate-url "{URL}"
fetch: url=https://web.archive.org/web/{TIMESTAMP}/{URL}, output_path={CASE_DIR}/research/archived/{filename}.md
Search the title in Google Scholar. Look for [PDF] links next to results — these are often author-posted copies on institutional servers.
search: query="{exact title}" site:scholar.google.com
For publications with metered paywalls (free article limits). Note: FT, Bloomberg, NYT, and WSJ are hard-paywalled — reader mode reveals nothing there; go to step 5 or 8 instead. For metered outlets:
Cmd-Shift-R or View > Show ReaderF9 or the reader icon in the address barWorks only on metered paywalls, not hard paywalls. This is a manual step — note in access_notes if reader mode was used.
Email the author directly. Most authors are willing to share their work:
Subject: Research access request — [{Title}]
I'm a journalist investigating [{topic}] and would like to access your paper "{Title}" (published {year}). Would you be able to share a copy? I'm happy to credit your work in my reporting.
Note the outreach in access_notes and set access_method to author_request_pending until a response is received.
Add access_method to every source entry in findings.json and fact-check.json:
{
"url": "https://example.com/paywalled-article",
"type": "news",
"accessed": "2026-03-15T14:20:00Z",
"access_method": "archive_copy",
"access_notes": "Retrieved via Wayback Machine snapshot from 2024-11-02"
}
Valid values for access_method:
full_text — Complete article accessed directly (no barrier)open_access — Legal open-access copy found via Unpaywall/CORE/authorfree_version — Freely distributed copy found via search (PDF, institutional mirror)archive_copy — Accessed via Wayback Machine or Archive.todaygoogle_scholar_pdf — PDF found via Google Scholar institutional linkreader_mode — Retrieved via browser reader mode on metered paywallauthor_provided — Author shared copy in response to requestauthor_request_pending — Author contact sent, awaiting responseabstract_only — Only abstract/summary availablepartial_text — Some content accessible (e.g., first paragraphs before hard paywall)cached_copy — Found in search engine cache (Google, Bing)institutional_access — Accessed through legitimate institutional subscriptionpreprint — Preprint version found (arXiv, SSRN, bioRxiv, etc.)inaccessible — Could not access after full hierarchy attempt| access_method | Confidence cap |
|---|---|
full_text / open_access / author_provided / institutional_access | No cap |
free_version / google_scholar_pdf / preprint | No cap (but note version date — may differ from published version) |
archive_copy / cached_copy | No cap (but note snapshot date) |
reader_mode / partial_text | medium at best — may be missing key content |
abstract_only | medium at best — abstract may omit key findings |
author_request_pending | low — pending, do not cite as verified |
inaccessible | low — cite the source but flag that content was not verified |
A finding that rests on an inaccessible source must say so explicitly in confidence_rationale. Do not inflate confidence because the source sounds authoritative if you have not read it.
inaccessible.access_notes so the next cycle doesn't repeat the effort.spotlight_safe.py shell-safety validation and {CASE_DIR} outputs, replacing upstream's Python library code.access_method vocabulary and confidence-cap table feeding findings.json/fact-check.json — no upstream equivalent.Reconciled against upstream @ 2097d218 (2026-07-06).
Adapted from claude-skills-journalism by Joe Amditis, released under MIT License.