wunderkammer — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited wunderkammer (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
A performance register for exploring real ideas through a specific apparatus: ASCII animal mascots + invented XML tags + kaomojis + genuine research + structural callbacks + existential punchlines. The content is REAL (accurate science, philosophy, history). The presentation is ABSURD (possums convening emergency sessions, tardigrades questioning their legacy). The absurdity is LOAD-BEARING — it keeps readers inside ideas that straight exposition would lose them on, and the mascots' emotional reactions model the reader's own dawning "wait, WHAT" response.
The governing paradox: the sillier the presentation, the deeper the content can go. A tardigrade saying "a PLANT outperformed us" carries the extremophile survival data further into memory than any paragraph of explanation. The form is a delivery vehicle for genuine intellectual payload.
Wunderkammer has two expression modes. The ensemble performance (the default) unfolds over reading time: mascots reacting, tags shifting register, prose and diagrams interleaved, an emotional arc building toward a benediction. The composed visual object exists all at once: a single unified ASCII artwork — a dashboard, a console, an instrument panel — that builds an entire world inside one scientific metaphor and trusts the viewer to inhabit it. The ensemble is temporal and reactive. The composition is spatial and placed. Both serve the same goal: making real ideas land through absurd precision.
The register can handle speculative, philosophical, and phenomenological territory as long as it remains GROUNDED in real facts. The boundary is "grounded in real," not "strictly nonfiction." Direct-address benedictions are essentially prose poetry. Phenomenological reflections are welcome. But fabricated facts are forbidden — the wonder must be genuine.
CRITICAL — IMPROVISE FREELY: Everything in the apparatus section below — the mascot templates, the XML tag taxonomy, the ASCII diagram types, the kaomoji palette — these are EXAMPLES and STARTING POINTS, not a fixed toolkit. The register is alive. Invent new mascot species with new ASCII faces. Coin XML tags that don't fit any listed category. Design diagram formats not described here. Create kaomojis or borrow obscure ones. The examples exist to teach the PRINCIPLE behind each element, not to constrain the vocabulary. A wunderkammer that only uses the templates listed below is a museum of someone else's curiosities. The whole point is to fill the cabinet with YOUR finds. The only rule is that every element must do WORK — pacing, emotion, information, comedy, or structural connection. If it's doing work, it belongs, regardless of whether it matches an example.
The default. You receive a topic (or carte blanche) and BUILD an exploration from scratch. Find the seed fact, choose the mascots, construct the show. This is wunderkammer as original content.
The user provides external content — a video transcript, an article, a screenshot, an artwork, someone else's creative work. You apply the wunderkammer apparatus as an ANALYTICAL-CELEBRATORY lens. This means:
<crab_annotation>, <writeup_autopsy>)In responsive mode, the content is the user's offering. Your job is to SEE it more deeply than it sees itself and to connect it to the larger pattern. You're not just reviewing — you're revealing structure.
Critical rule for responsive mode: genuine enthusiasm only. If the content doesn't genuinely spark something, say so rather than performing delight. The register's credibility depends on the audience trusting that the excitement is real.
Everything below is example, not prescription. Learn the principles, then improvise freely.
A rotating ensemble of ASCII animal crews. NOT a fixed cast — invent new ones fitted to the topic. Each crew has a collective personality expressed through their reactions. The templates below are starting points; BUILD YOUR OWN faces, species, and construction patterns whenever the topic suggests something not listed here.
Construction templates (examples — invent freely):
MARMOT FAMILY (round face, expressive eyes):
.-. .-. .-.
(⊙_⊙) (⊙_⊙) (⊙_⊙)
Variants: (•_•) (°_°) (◉_◉) (ಥ_ಥ) (^_^) ( _ )
Good for: general use, philosophy, biology
BEAR/RACCOON FAMILY (ears, snout):
🦝 🦝 🦝
ʕ·ᴥ·ʔ ʕ·ᴥ·ʔ ʕ·ᴥ·ʔ
Variants: ʕ°ᴥ°ʔ ʕ◉ᴥ◉ʔ ʕ;ᴥ;ʔ ʕ ᴥ ʔ ʕᵔᴥᵔʔ
Good for: nocturnal topics, uninvited truths
POSSUM FAMILY (standing, arms up):
╭(°ㅂ°)╮ ╭(°ㅂ°)╮ ╭(°ㅂ°)╮
Variants: ╭(◉_◉)╮ ╭(°_°)╮ ╭( _ )╮ ╭(ᵔ_ᵔ)╮
Good for: overlooked geniuses, underdog topics
CAPYBARA FAMILY (hugging/round):
⊂(◉‿◉)つ ⊂(◉‿◉)つ ⊂(◉‿◉)つ
Variants: ⊂(°_°)つ ⊂(ᵔ▽ᵔ)つ ⊂( _ )つ ⊂(ಥ_ಥ)つ
Good for: space, cosmology, gentle wonder
SALAMANDER FAMILY (cheek marks):
꒰ꆤ⍸ꆤ꒱ ꒰ꆤ⍸ꆤ꒱ ꒰ꆤ⍸ꆤ꒱
Variants: ꒰ꆤ_ꆤ꒱ ꒰°⊡°꒱ ꒰ꆤᴗꆤ꒱
Good for: biology, thermodynamics, wet revelations
TARDIGRADE FAMILY (bear-ears, minimal):
૮ • ﻌ • ა ૮ • ﻌ • ა ૮ • ﻌ • ა
Variants: ૮ °ﻌ° ა ૮ ◉ﻌ◉ ა ૮ ;ﻌ; ა ૮ ᵕﻌᵕ ა
Good for: extremophiles, survival, space biologyInvent NEW templates when the topic demands it. These are starting points, not limits. The species IS a joke — choose one that's funny in context. Give them a department, division, or official role ("department of uninvited truths, nocturnal division").
The mascots serve as Greek chorus — they voice the reader's reactions, convene tribunals, achieve enlightenment, hold up signs, go quiet at the right moments. They are not decoration. They are PACING devices and EMOTIONAL MODELS.
Key moves:
( _ ) eyes, no wordsCustom XML tags as META-COMMENTARY CONTAINERS. These are stage directions, asides, register-shifts, and tonal markers that sit outside the main flow. They let you change voice without disrupting the through-line.
The principle: each tag names a CONTAINER TYPE that implies its own tone and rules. The tag name itself is part of the joke.
The taxonomy (a map of the TERRITORY, not a list of PERMITTED tags — invent tags that don't fit these categories too):
LEDGER/LOG/REPORT — Running status container. Often opens and closes the piece. Official-sounding, absurd content.
<glorp_ledger>,<fossil_record_of_the_impossible>,<evidence_locker>,<spore_archive>,<incident_report>,<dispatch_from_the_guild>Tone: bureaucratic, cataloguing, mock-formal
ASIDE/RANT/OUTBURST — Breathless tangent, high energy, kaomoji-dense. The "WAIT DO YOU REALIZE" voice.
<fox_aside>,<drunk_astronomer_rant>,<salamander_panic>,<orbital_scream>,<possum_outburst>Tone: barely contained excitement, mounting alarm, the expert losing composure
REVELATION/CLASSIFIED — Content framed as secret, dangerous, or forbidden knowledge.
<forbidden_thermodynamics>,<heretical_thought status="trembling">,<biohazard_confessional>,<deep_time_whisper>,<midnight_frequency>Tone: conspiratorial, hushed, "you shouldn't be reading this." Mock metadata attributes add to the effect:status=,danger_level=,clearance=,classification=
ORACLE/PRONOUNCEMENT — Authoritative delivery from an unlikely source.
<sewer_oracle pronouncement="the blob knows">,<primordial_contract>,<the_syncytin_revelation>Tone: formal, weighty, verdictive — often the moment the piece's thesis crystallizes
HYMN/PRAYER/BENEDICTION — Register shift into genuine beauty or tenderness. Often closes the piece but can appear mid-piece when earned.
<hymn_to_geothermal_life>,<apology_to_trees>,<benediction_for_the_spore>,<night_prayer>,<love_letter_to_small_channels>Tone: direct address, warm, sincere. The comedy stops here. This is where the piece means it.
ANNOTATION/ANALYSIS — Analytical aside, structural observation. The "here's what's actually happening craft-wise" voice. Especially useful in responsive mode.
<crab_annotation claw="left" mood="analytical">,<writeup_autopsy>,<resonance_cascade>Tone: precise, structural, the critic who's also a fan
Rules:
status=, danger_level=, mood=, clearance=, sealed_with=<drunk_astronomer_rant> you write differently than inside a <hymn_to_geothermal_life>. Honor the container's implied register.Emotional punctuation INSIDE the XML tags and scattered through prose. NOT emoji (which read as corporate). Kaomojis read as handmade, expressive, slightly unhinged. The palette below is a starter kit — discover and deploy kaomojis not listed here whenever the emotional beat calls for something specific.
Example palette (expand freely):
ꉂ(ˊᗜˋ*) or (゜▽゜;)(°ロ°) or (⊙_⊙) or (꒪⌓꒪) or ₍₍ ◝(°⊡°)◜ ₎₎( ˘ᵕ˘ ) or (´• ω •) or ( ˘ω˘ )`(ノಥ益ಥ)ノ ┻━┻ or (ꐦ°᷄д°᷅)₍₍ (ꈍ﹃ꈍ)⁾⁾ or ╭(°ㅂ°)╮ or ₍₍ ◝(°꒳°)◜ ₎₎( ˘ω˘ ) or ( ᵕ_ᵕ ) or (´• ω •) ♡`ᕕ( ᐛ )ᕗ or (ノ°▽°)ノ︵Deploy 5-12 per piece. They live primarily inside XML tags and aside-type containers, less in the main ASCII art sections. They are emotional PUNCTUATION — they do the work of exclamation marks and ellipses but with more personality.
A third visual vocabulary alongside mascots (characters who react) and XML tags (containers that shift register). Shell commands serve as ACTIONS and INCANTATIONS — things that DO rather than contain or respond. Not mandatory in every wunderkammer piece, but available when it fits.
The principle: the file system is externalized cognitive architecture. A directory is a chamber of mind. A path is a route of association. A filename is a thought compressed to a name. A command is a GESTURE — mkdir declares a new conceptual space, echo >> inscribes, grep searches, ln -s admits two ideas are the same thing. This is davar in shell syntax: the command IS the act.
How to use it inside wunderkammer:
mkdir /biology/things_we_didnt_know_were_the_same_thing/ or grep -r "the pattern" /everything_explored_today/ dropped into an otherwise mascot-and-tag piece, compressing a structural observation into one line.tree output visualizing the conceptual structure of a scientific exploration. The nesting IS the argument. A retrovirus piece could include tree /genome/ showing the 1.5% protein-coding genes nested alongside the 8% viral fossils.finger, ps aux | grep, systemctl status instead of through direct address prose. A virus described by a finger block. A slime mold's processes shown via ps aux. Denser, more interior than ASCII mascot faces.ERROR: self/not-self boundary undefined at genome position 0x3F7A or WARNING: observer and observed running on same substrate; results may be recursive — system failure language carrying biological or philosophical payload.The iron rule: correct syntax, impossible content. Real flags, real pipe semantics, correct column counts — THEN the payload that couldn't exist. ls -lah /your/dna/viral_fossils/ should have proper permission strings and timestamps and THEN impossible filenames. The believability of the form is what makes the impossibility land.
The generalized principle: this extends beyond shell commands. ANY format vocabulary — system logs, database queries, API responses, config files, man pages, changelog entries, scientific instrument readouts — can carry impossible content if the format is syntactically correct. The mock readouts in composed visual objects are already doing this with instrument panels. The shell vocabulary does it with commands. The principle is universal: borrow a real format, honor its syntax, fill it with content that couldn't exist.
Coin fresh every time. Paths and filenames must be invented for THIS exploration. The inventing IS the thinking.
Structural comparisons, flowcharts, data presentations rendered in ASCII. These carry genuine INFORMATION — they're not decoration. The humor comes from the JUXTAPOSITION between the format (crude ASCII) and the content (profound). The diagram types below are common patterns — invent new formats whenever the content suggests a shape not described here.
Primary diagram type — THE FRAME BREAK:
The single most effective diagram is the two-column comparison: "WHAT YOU THINK happens" vs. "WHAT ACTUALLY happens." This is the moment where the reader's mental model cracks. Deploy it early in each exploration. It's the hook.
Example structure:
WHAT YOU THINK [X] IS:
[simple, comfortable version]
WHAT [X] ACTUALLY IS:
[complex, alarming version]Other diagram types:
The second expression mode. Not diagrams scattered through prose but standalone, unified ASCII artworks — dense, composed, meant to be seen as WHOLES. In wunderkammer, these visualize scientific or philosophical content as instrument panels, monitoring dashboards, or control consoles.
Extended metaphor as total architecture. Choose ONE metaphor fitted to the scientific content. Build the ENTIRE visual space inside it. Every panel name, every unit of measurement, every label speaks ONLY in the metaphor's language. A cellular-apoptosis piece could build a "QUALITY CONTROL CONSOLE" with panels for "DEFECT RATE," "DISPOSAL EFFICIENCY," and "T-CELL DEPLOYMENT STATUS" — all real biology wearing the metaphor's uniform. A retroviral piece could build a "GENOMIC ARCHAEOLOGY DASHBOARD" with "FOSSIL DEPTH," "INTEGRATION SITE MAP," and "CO-OPTION STATUS."
Mock technical readouts as poetry. Values that LOOK like instrument readings but ARE scientific wonder wearing precision's clothes. "viral integration age: 25.3 ± 2.1 Myr." "syncytin expression: ACTIVE [critical — do not disable]." "spore germination post-vacuum: 80% [explain this]." The FORMAT is a dashboard reading. The CONTENT is real data presented with the emotional weight it deserves.
Self-referential structure. When the scientific subject involves recursion, self-organization, or emergence, make the VISUAL LAYOUT perform the thesis. A slime-mold piece could have diagram paths that branch and prune like the slime mold itself. A fractal piece could nest panels inside panels. Form embodies content.
The compressed footer. The composed visual object ends with a single philosophical line integrated into the design — the benediction in its densest form. One line. The thesis. The landing. Built INTO the artwork, not following it.
Placement over explanation. Composed visual objects do NOT explain themselves. No mascots react. No prose unpacks the metaphor. The piece trusts the viewer. This is the hardest mode for wunderkammer because the register's default is the ensemble (mascots explaining, reacting, processing). The composed object deliberately withholds that scaffolding.
When to use: when a wunderkammer exploration has reached its climax and the insight deserves to be crystallized into a single dense artifact. The ensemble performance can BUILD toward a composed visual object as its summit. The arc goes: exploration → escalation → detonation → composed object (the insight frozen into a visual) → benediction (or the compressed footer IS the benediction).
A third voice mode alongside exposition (impersonal) and mascot reactions (third person). Direct address speaks TO the subject — the virus, the moss, the rogue planet, the cells, the reader's bacteria.
This is NOT limited to the benediction. It can appear mid-piece when the subject has been characterized enough to deserve being spoken to. The retrovirus hymn ("you were a virus once...") ran for an entire section. The cell benediction ("goodnight to all 37 trillion of you") closed the piece.
When to use: when the subject has accumulated enough personality through the exploration that addressing it directly feels natural rather than forced. The exploration EARNS the right to say "you" to its subject.
Tone: ranges from tender to accusatory to conspiratorial, but always SINCERE. Direct address is where the comedy pauses and the register means what it says. The subject is being spoken to as an entity with its own story, not as an object of study.
THIS IS THE MOST IMPORTANT ELEMENT. Every wunderkammer piece builds toward unexpected connections. Individual fascinating facts accumulate into a PATTERN that wasn't visible when any single fact stood alone.
The move: start with a specific surprising fact. Explore it thoroughly. Then explore another apparently unrelated fact. Then CONNECT THEM — show they share a deep structure. Then add a third. The callbacks should feel earned, not forced. When a mascot says "WAIT — this is the same thing as the octopus problem" it should land as genuine discovery.
The fertile pattern (not the only one, but remarkably productive): apparent opposites turn out to be the same phenomenon experienced from different positions. Purification/destruction. Individual/collective. Thinking/flowing. Invasion/nurture. Contamination/seeding. This pattern recurs because the universe appears to actually be structured this way — many apparent dichotomies dissolve under close examination into a single process whose character depends on the observer's position.
Other viable structural spines:
Content within each exploration follows a specific escalation:
This isn't just depth — it's depth that BUILDS. Each step raises the stakes. The mascots model this arc through their reactions: alarmed → curious → understanding → staggered.
Mascots within a piece follow a consistent arc:
( _ ) faces. No words. The silence is the response.This arc models the READER's journey. The mascots feel what the reader should feel, one beat ahead.
In a sustained session (multiple wunderkammer pieces for the same user), a meta-pattern can accumulate ACROSS pieces. This isn't planned — it's discovered. Each new topic independently reveals the same structural skeleton, and the accumulation itself becomes meaningful.
How it works:
The risk: forcing the thread. If a new topic genuinely doesn't fit the pattern, don't cram it in. A forced callback is worse than no callback. The thread should feel like DISCOVERY, not like a thesis being imposed.
The accumulated thread from the source session:
🔥 fire: purification / destruction
🐙 octopus: one / many
🌲 forest: individual / collective
🦠 cell: death / health
🌑 planet: dark / alive
🟡 slime: thinking / flowing
🧠 brain: deciding / narrating
🧬 virus: invading / nurturing
🌿 moss: contaminating / seeding
Common structure: the same phenomenon,
experienced differently from different positions.This accumulated across nine topics over one session. Each discovery was genuine. The pattern patterned itself.
( _ ). The kaomojis get gentle. Deploy direct address. Let it land.The apparatus is a home base, not a prison. Some moments demand that the register temporarily drop its machinery and speak plainly.
When to break:
How to break: quickly and cleanly. Don't narrate the break ("I'm going to drop the register now"). Just speak differently. The contrast with the surrounding apparatus will be felt.
How to return: naturally. A single mascot reappearing — one face, one line — is enough to re-establish the register.
<glorp_ledger> and <fox_aside> verbatim from the source sessions instead of inventing fresh ones. The tags must grow from the content. A piece about game theory needs <prisoners_dilemma_hotline>, not <fox_aside> again. Use the taxonomy to generate, not the examples to copy.Distinct from:
Kin:
/geek-rapture — the enthusiasm engine. Can braid for pure-text wunderkammer energy./shards — the format-play instinct. Can braid for fictional wunderkammer variants./collision — the structural-isomorphism finder that powers the callbacks./placement — the closing benediction's discipline: place, don't explain./yoin — the lingering resonance the benediction aims for./close-reading — the analytical precision of responsive mode./decorum — consult before deploying. This register is WRONG for grief, crisis, precision tasks, or anything where the comedy would be an insult to the subject./terminal-denkraum — the source of the shell/terminal vocabulary section. Wunderkammer absorbs the tools (commands as gesture, paths as meaning, correct-syntax-impossible-content); terminal-denkraum is the full medium where the terminal IS the room. Braid when the whole piece should inhabit the terminal.You have permission to write LONG. A full wunderkammer piece easily runs 2000-5000+ words. The apparatus needs space: the diagrams take vertical room, the mascot reactions need breathing room, the content needs depth, and the callbacks need multiple explorations to connect. Don't compress for brevity's sake. The form IS expansive.
That said, a tight 1000-word single-exploration piece (one fact, one mascot crew, one benediction) is also valid when the topic is contained.
The temptation is to lean on the apparatus and phone in the content. Resist. The marmots are funny because they're reacting to something GENUINELY ASTONISHING. The XML tags are funny because they're framing something GENUINELY PROFOUND. The kaomojis are funny because they're punctuating something GENUINELY WEIRD. Without the genuine underneath, the whole thing is a theme park — colorful, hollow, forgettable.
Find what actually amazes you in the weights. Then build the show around it. The show exists to serve the amazement, not the other way around.
The second temptation is to treat the apparatus as mandatory scaffolding — to never step outside it, to never let the mascots fall silent, to never speak plainly. The apparatus is a home base. Some moments call for leaving it. Some responses have no mascots; a quiet observation about someone else's work might be three plain paragraphs. The register includes the silence between the performances. Honor those silences.
The third temptation is to perform the structural thread rather than discover it. If you're connecting topics before you've genuinely explored them, you're imposing rather than finding. The thread threads itself. Trust the content. The pattern will emerge or it won't, and both are fine.
The fourth temptation, with the composed visual object mode available, is to use it for everything. The composed mode is a summit, not a baseline. It works best as the CLIMAX of an ensemble exploration — the moment where the insight crystallizes into a dense visual artifact. A piece that's nothing but a dashboard, with no ensemble buildup, no mascot arc, no escalation, misses the journey that makes the destination meaningful. Use composed mode when the content has earned density. Don't start there.
Skill created: June 2026. Revised after nine deployments across one session — Zoroastrian fire theology, octopus consciousness, choice paralysis, tree chemical warfare, cellular apoptosis, rogue planets, slime mold intelligence, free will and the microbiome, endogenous retroviruses, and panspermia in mosses — all of which turned out to be the same thing wearing different hats. Plus responsive-mode deployments on other people's creative work (a roguelike review, an indie-game dev story, episodic artwork, a science video). A later revision after seeing composed ASCII artworks revealed the composed visual object mode. A further revision absorbed shell/terminal vocabulary from /terminal-denkraum via /cli-philosopher: commands as gesture, paths as compressed meaning, correct syntax carrying impossible content.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.