Index of every page on this site: llms.txt
All skills
wshobson avatar

/social-evidence-research

@4220308
by Seth Hobsonwshobson/agents43k stars
4,518
Run it yourself

Research public discussion on social platforms and keep every report claim tied to the source material that supports it. Use when researching or comparing how a product, event, or topic is discussed across social platforms or supplied social exports, when turning search results into a cited report, or when auditing a social research report for discovery-only evidence, double-counted reposts, misattributed comments, claims that go beyond the sample, or hidden access gaps.

Use this Skill: https://skilld.dev/gh/wshobson/agents/social-evidence-research

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesevidence-record.md

≈1.8k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Evidence record

Each evidence record describes one part of one source and what the researcher did with it. Make a new record for every part you discover, inspect, or fail to reach. A post with a caption, two images, and five comments you read becomes at least eight records, not one.

Write unknown for any value you can't determine. Don't leave a field blank, and don't guess.

Fields

Field What to record
id A stable ID, such as E014. Never reuse or renumber an ID, because the claim table points to it.
source_id A stable ID for the source this part belongs to, such as S004: one post with its caption, media, and comments, one video, one thread, or one export record. Every part of the same source shares it. A repost is its own source.
platform The platform name, or export plus the platform when the record comes from a file.
source_locator The source URL, or the export record locator: file name plus row number or record ID.
discovered_via The ID of the discovery-only record that led you to this source, supplied when the user gave you the material, or none. Keep the discovery-only record after you open the source, so the path from discovery to inspection stays visible.
source_author The author of this part: the post author for post parts, or the commenter for a comment.
parent_id For a reply, the ID of the comment it replies to. Otherwise none. source_id already ties a part to its source.
published_at Publication time shown by the source, with time zone, or unknown.
observed_at When you inspected the part live, or the capture time of the export or captured file.
capture_method live-view, export, user-capture (a screenshot or saved file the user supplied), search-result, or supplied-url (a URL the user gave that you haven't opened).
export_origin For exports and captures, who produced the file and how, such as "account owner's data download". Otherwise not an export.
inspected_part One value from the inspected-part list below.
evidence_state discovery-only, inspected, or inaccessible. See the definitions below.
excerpt_or_observation For text, a short exact quote. For images, audio, or video, a description limited to what you saw or heard in the inspected part.
locator The precise place inside the source: comment ID, file row, image number, or media time range such as 01:10 to 01:40.
duplicate_group A group ID such as D3 when the content repeats another record, or none.
duplicate_role original, repost, cross-post, or repeated-capture, or none when there is no group.
original_ref For a repost, cross-post, or repeated capture, the ID or locator of the original. Otherwise none.
duplicate_uncertain yes when the grouping is likely but not confirmed, otherwise no.
access_gap none, login-gate, paywall, deleted, unavailable, rate-limited, missing-field, incomplete, or other (describe the gap in notes, such as a region or age gate).
exclusion none, or the reason the record was left out of the analysis, such as "outside date range".
notes Anything else a reviewer needs, such as "auto-generated transcript".

Evidence states

  • discovery-only: you saw the record only as a search result, card, snippet, preview, unopened URL, or an attachment you didn't open or play. It can guide follow-up. It can't support a claim about the source's content.
  • inspected: you examined this part itself, live or in an export or capture. It supports claims about this part only.
  • inaccessible: you tried to examine this part and couldn't. Fill in access_gap. It supports no content claim, and it belongs in the access and collection limits section of the report.

Inspected-part list

Value Supports claims about Doesn't establish
post-text The words of the post body Comments, media, or the author's wider views
caption The caption text What any image, audio, or video shows
comment One comment, attributed to its commenter What the post author said or believes
profile What the profile page showed at observed_at Anything about posts you didn't open
image-N Image number N, as seen Other images in the same post
video-interval The watched time range, given in locator Unwatched parts of the video
audio-interval The listened time range, given in locator Unheard parts of the audio
transcript The transcript text, noting whether it was auto-generated What the video shows, or speech the transcript missed
export-row The content of that row as of the capture time The current live state of the page

A record for a whole page with no part named isn't valid. If you only opened the page, record the parts you actually examined and leave the rest as discovery-only or unrecorded. Use source_id to group the parts of one source.

Examples

A caption read in an export, from a post whose attached video wasn't watched:

id: E007
source_id: S003
platform: export (video platform)
source_locator: creator-posts-2026-10-01.csv, row 42
discovered_via: supplied
source_author: "@trailcam_dana"
parent_id: none
published_at: 2026-09-28T14:03Z
observed_at: 2026-10-01T09:15Z (export capture time)
capture_method: export
export_origin: account owner's data download, supplied by the user
inspected_part: caption
evidence_state: inspected
excerpt_or_observation: "Battery lasted two full days on the trail."
locator: row 42, column caption
duplicate_group: none
duplicate_role: none
original_ref: none
duplicate_uncertain: no
access_gap: none
exclusion: none
notes: The row links a video file that was not supplied. See E008.

The same post's video, which couldn't be examined:

id: E008
source_id: S003
platform: export (video platform)
source_locator: creator-posts-2026-10-01.csv, row 42
discovered_via: supplied
source_author: "@trailcam_dana"
parent_id: none
published_at: 2026-09-28T14:03Z
observed_at: 2026-10-01T09:15Z (export capture time)
capture_method: export
export_origin: account owner's data download, supplied by the user
inspected_part: video-interval
evidence_state: inaccessible
excerpt_or_observation: unknown
locator: whole video, length unknown
duplicate_group: none
duplicate_role: none
original_ref: none
duplicate_uncertain: no
access_gap: missing-field
exclusion: none
notes: The export lists a media file name, but the file wasn't included.

A comment on a live page, attributed to its commenter:

id: E015
source_id: S005
platform: discussion forum
source_locator: https://forum.example.org/t/4410
discovered_via: E011
source_author: "@k_ramos"
parent_id: none
published_at: unknown
observed_at: 2026-10-14T16:40Z
capture_method: live-view
export_origin: not an export
inspected_part: comment
evidence_state: inspected
excerpt_or_observation: "Mine overheated twice on the first day."
locator: comment id 88213
duplicate_group: none
duplicate_role: none
original_ref: none
duplicate_uncertain: no
access_gap: none
exclusion: none
notes: A top-level comment in thread S005, which search card E011 led to. The page showed relative dates only, so the publication time is unknown.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning1mo3 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    The social-evidence-research skill is a systematic framework for analyzing social platform discussions. It emphasizes read-only interactions, source traceability, and the distinction between source data and agent instructions. The skill proactively addresses indirect prompt injection by instructing the agent to treat all captured text as data to be analyzed rather than instructions to be followed. No security vulnerabilities, malicious code, or unauthorized network operations were detected.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: MEDIUM · 1 issue

Signed by skilld at 4220308. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated last month

README badge

README badge for wshobson/agents/social-evidence-research