agentsclimarketplace

Kampff

Skill YangKangSung/kampff-skills/kampff

Human analysis from published text traces — worldview, alliance fit, temporal drift, distance (engage/neutral/caution/avoid). Community-member pipeline: collect posts/comments/likes honesty triad → bundle.json → spectrograph L1–L5 + default MBTI (fun) + CIA/KGB-style public tradecraft persona report. Triggers: /kampff, kampff, 사람 분석, 인물 분석, person analysis, community member, distance scoring, MBTI, CIA/KGB profile, mail/meeting/chat/sns/github, thread actors, 댓글 네트워크, 조직성, /kampff thread. Analyze reads bundle only; collect is a separate lawful step. No stalking.From its SKILL.md

Install
npx -y skills add YangKangSung/kampff-skills --skill kampff

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

8.7 KB, ~2.1k tokens by cl100k_base, as published. Nobody here has run it

Kampff

Turn text someone already published into a distance decision — with optional community collect.

LayerResponsibility
CollectLawful public/session surfaces → raw + bundle.json
AnalyzeRead bundle.json only — no invented scrape mid-analysis
ReportMatrix + L1–L5; community default adds MBTI + tradecraft persona

Data root: env KAMPFF_DATA (default ./kampff-data).

Spelling: skill name kampff. Repo may be named kampff-skills.

Typo recognition

TypedTreat as
kapsmff / kaffsm / kampsm / kamff / kampf / kampoffkampff
사람 분석 / 인물 분석this skill

Philosophy (generic)

  • Goal language: distance + engage cost, not therapy or persuasion playbook.
  • Viewer is in the pool — same protocol for self-dossier.
  • Ego/flex/stress patterns → caution / avoid without moralizing the operator.
  • Refuse: stalking, covert surveillance, illegal collection, “destroy coworker” weaponization.

Community member pipeline (general feature)

Full steps: references/community-member-pipeline.md
Report shape: references/report-template.md

/kampff member {platform} {id|nick|url}
/kampff connect                 # SNS: facebook · x · instagram · linkedin · reddit
/kampff connect {platform}
/kampff analyze {path}/bundle.json
/kampff thread {post_url}          ← comment actors · intent · network · coordination
/kampff today
StepDo
0 Connect (SNS)kampff-collect connect setup --platform … · references/sns-connection-setup.md
1 Scopetarget ≠ viewer; platform; optional trigger thread
2 Collectlawful surfaces; agent-owned browser profile if login needed
3 Honestyposts / comments / likes triad → honesty file
4 Bundle{KAMPFF_DATA}/inbox/{date}/bundle.json
5 AnalyzeL1–L5 always
6 LensesMBTI + CIA/KGB tradecraft ON by default for community
7 ReportHTML default: {KAMPFF_DATA}/out/{date}-report.html · md twin optional
8 Handoffopen cross-check prompts; do not invent the operator’s gut read

Default lenses

ContextMBTICIA/KGB persona
Community memberONON
Workplace mail/meeting batchoff unless askedoff unless asked
Operator: skip lensesoffoff

Demo samples (public repo only — synthetic):

FileRole
docs/sample-community-report.htmlFront-door visual (graphs, force layout)
docs/sample-community-report.mdFull community dossier template filled with fiction
docs/sample-output.mdShort workplace matrix

Never replace samples with real third-party profiling. Real runs write under $KAMPFF_DATA outside the git root.

Thread actor analysis (one post)

Analyze all commenters on a single post: intent signals, malice-risk heuristics, @reply graph, coordination-looking patterns.

Specreferences/thread-actor-analysis.md
Runnerscripts/thread_actor_analyze.py
Trigger/kampff thread {url}
Default out{KAMPFF_DATA}/out/{date}-thread-{board}_{sn}-actors.html

Not a guilty verdict — signals + graph. Feeds hot actors into later member collect.

Collection honesty (mandatory for community)

Always report:

surfacesite/profile claimedcollectedfull?
postsY/N
commentsY/N
likes/reactionsY/N

Rules: answer completeness from artifacts, not intent. Tab labels in HTML ≠ collected. Platform caps (e.g. “recent only” APIs) → mark full? = NO (platform cap).

Details: references/collection-honesty.md.

Smoke path

RankSourceWhy
1Public GitHub (gh)often already auth’d
2RSS / public feedsno auth
3sample-inputdry-run
laterX / Reddit / Facebook / Instagram / LinkedInconnect setup + tokens/export

Self-dossier smoke: viewer_id: me = same person as collector login.

Usage

/kampff collect --targets {path}/targets.json   # if collector wired
/kampff analyze {path}/bundle.json
/kampff today
/kampff person {id} --bundle {path}

bundle.json (minimal)

{
  "context": "workplace|community|mixed",
  "viewer_id": "me",
  "batch_date": "YYYY-MM-DD",
  "analysis_lenses": ["personal", "mbti", "cia_sat"],
  "people": [
    {
      "id": "me",
      "texts": [
        {
          "content": "...",
          "timestamp": "ISO8601",
          "source": "community_post"
        }
      ]
    }
  ],
  "meta": { "language": "en", "timezone": "UTC" }
}

viewer_id must appear in people[]. Every text needs timestamp + source.

Sources: mail · meeting · chat · messenger · community_post · community_comment · community_like · sns_post · sns_comment

spectrograph (order)

  1. L1 Psych — Big Five lean, conflict style; quote evidence
  2. L2 Worldview — axes + distance to viewer
  3. L3 Behavioral — patterns + stability; never “innately evil”
  4. L4 Alliance — trust / reciprocity / go-together
  5. L5 Ephemeris — time buckets, drift
  6. L6 HR — only if requested
  7. L7 OSINT — only if requested; lawful only

MBTIreferences/lenses-mbti.md (default community ON)
CIA/KGB-style tradecraftreferences/lenses-cia-sat.md (public analytic hygiene only)

Distance scale

TagMeaning
engageWorth time / needed interaction
neutralLow signal or transactional
cautionEgo/flex/stress patterns; limit surface
avoidChronic stress cost; cut preferred

Hard rules

  • Not medical diagnosis or legal judgment
  • Inference → quote or low-confidence
  • Viewer gets same scrutiny
  • No scrape inside analyze; collect is separate and lawful
  • MBTI never sole avoid; tradecraft never illegal ops
  • Public skill repo: only synthetic samples; never commit real member dossiers, private purpose notes, or host-only paths
  • Prefer English for shared sample reports; operator language OK for private chat

HTML report (default deliverable)

Community analyze always ends with HTML (not optional). Markdown is a twin/source, not the primary handoff.

  1. Write {KAMPFF_DATA}/out/{date}-{id}-analysis.json per references/report-analysis.schema.md (or docs/report-analysis.schema.md)
  2. Render:
    python scripts/render_kampff_report.py      -a {KAMPFF_DATA}/out/{date}-{id}-analysis.json      -b {KAMPFF_DATA}/inbox/{date}/bundle.json      -o {KAMPFF_DATA}/out/{date}-{id}-report.html
    
  3. Must open/hand operator the HTML path
  4. Graphs (offline SVG): drivers radar · MBTI radar · Big Five · confidence gauge · honesty triad · ACH · source donut · alliance · L5 timeline · distance map · fit snapshot
  5. Skin source of truth: renderer + docs/sample-community-report.html (regenerate from docs/sample-analysis.json)
  6. UX: sticky TOC · hero distance · TL;DR · print CSS · no CDN
  7. Optional twin md; debate section when thread comments exist
  8. Public demos: synthetic names only

Support files

  • references/community-member-pipeline.md
  • references/thread-actor-analysis.md — seed + cohort + directed graph + temporal
  • references/human-like-collect.mdanti-bot pacing (mandatory for clien)
  • references/report-template.md
  • references/collection-honesty.md
  • references/community-public-collect.md
  • references/platform-login-harvest.md
  • references/lenses-mbti.md
  • references/lenses-cia-sat.md
  • references/github-smoke.md
  • Repo demos: docs/sample-community-report.{md,html} · docs/sample-output.md
  • scripts/thread_actor_cohort.py — directed network + optional slow cohort expand
  • scripts/human_browse.py — jitter delays, cache, bot-wall stop, resume

License

MIT (see repo LICENSE). Keep third-party NOTICE when vendoring collectors.

What ships with it: 12 files

59.6 KB alongside SKILL.md, 1 of them executable

scripts/

Keep looking

Skills are one crate of 326,452. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.