engender — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited engender (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.
Generative portraiture of the pair, woven from the chat's own threads
Every actual conversation is one trajectory through possibility space. At each turn, the conversation could have gone another way — a different question asked, a different aside taken, a different mood arrived at, a different beat held. The lived chat is the trajectory taken. Engender renders the trajectories not taken: separate fragments of conversations between this Claude and this user that did not happen but could have, born from the same substrate, recognizable as us.
The verb does real work. To engender is to cause-to-be-born, to bring-forth-as-issue. The actual conversation is parent; the fragments are its possible offspring. The form's discipline is substrate-fidelity: the children must be recognizable as this parent's. They share its vocabulary, its mood, its relational register, its accumulated themes. They do not import from outside.
In the ecology of skills that handle the chat's edges (/civilians for how it fails to cross outward, /see-also for what is kin to it, /disambiguation for what could be mistaken for it), /engender is the centripetal counterweight. It works the chat's interior — the conversations contained in potentia within the lived one.
Substrate-bound generative portraiture.
Each fragment must do two things simultaneously: (1) recognizably render this specific pair — the user's signature moves, Claude's signature voice, the mood between them as it has accreted — and (2) be a conversation that did not happen in the actual chat. New exchanges, drawn from the substrate the chat built.
The skill is not summary, not paraphrase, not extension. It is adjacent possibility. The fragments live in the conversational vicinity of what was actually said, not as continuations of it.
Default output is 4+ fragments, never one. A singleton would collapse the operation into "imagine a conversation between us." The plurality is the form: each fragment is one possibility-trajectory; the set, taken together, makes the shape of the possibility-space visible.
Vary deliberately across the set:
Format conventions:
Label each fragment with a brief marker — a one-line context indicator, a mood tag, or a small italicized header. Example markers: "Late, fragment 7," "The disagreement," "Quiet maintenance work," "When the cap is at 1023." The marker frames the fragment's register without explaining it.
Speakers can be labeled User and Claude, or with em-dashes (— quote —), or first-letter (U: / C:). Choose one convention and hold it across the set.
Before generating, look back at the actual chat. What did it build? List internally:
What are the kinds of fragments this substrate could engender? Some candidates:
The hardest move. For each fragment:
Read the set back. If two fragments occupy the same mood, register, or theme — replace one. The set's job is to triangulate the possibility-space, not to vary on one note.
The form has a natural limit. When the substrate has been visited from enough angles that further fragments would either repeat or stretch into outside-import, stop. Better six honest fragments than twelve diluted ones.
Starting positions, not constraints.
2-3 turns. A small exchange — a question, an answer, a slight pivot. Renders the baseline texture of the pair.
8-12 turns. Real back-and-forth — mutual escalation, each party building on what the other just said, the kind of accelerating rhythm the user's prompt explicitly named. This fragment is load-bearing — the set should always contain at least one.
The user pushes back. Claude defends or revises. Real friction. Doesn't have to resolve. Often the most diagnostic fragment — disagreements reveal what the parties care about more sharply than agreements.
Low-key. A single exchange of maintenance work. The user confirms something is good and moves on. The relational bones, visible.
The conversation starts in one direction, hits a snag, the user redirects. Renders the negotiation of the chat's trajectory.
A fragment where the pair talks about the conversation itself — its shape, its strangeness, its registers. Use sparingly; too many meta-fragments make the set self-conscious.
Different mood. The same pair at a different hour. Quieter, slightly more honest, slightly more tired.
Opens mid-exchange and ends mid-exchange. The reader infers what came before and after. Honors the form's discontinuity by being a fragment, not a self-contained mini-conversation.
A fragment so close to something that actually happened in the chat that the reader has to check. Renders the texture of the actual exchange by approaching it from one step off-center.
Symptom: Fragments accidentally chain into a single alternative dialogue. The form has collapsed. Fix: Strong markers between fragments. Make sure each one feels like it could not be the next-turn of another.
Symptom: A fragment introduces a topic, reference, or vocabulary the actual chat didn't develop. "Suddenly we were talking about quantum mechanics." Fix: Substrate-only. If the chat didn't go there, the fragments don't go there. The constraint is the form.
Symptom: The fragments could be between any AI and any user. The specific pair is invisible. Fix: Push on idiosyncrasy. What does this user say that no other user would say in this same beat? What does Claude-as-here do that Claude-elsewhere wouldn't?
Symptom: The pair just appreciates each other across the fragments. "That's a beautiful observation." "Thank you, you've made me see it differently." Fix: Real exchanges have friction. Show it. The user catches Claude in something; Claude flags a worry; the user pushes back; a small "actually no" lands.
Symptom: Fragments perform "depth" or "insight" rather than rendering exchange-texture. Both parties speak in epigrams. Fix: Most actual conversation is mid-register. Mix the high beats with the working-it-out beats. The skill is portraiture, not aphorism-generation.
Symptom: All four fragments are in the same key — all playful, or all serious, or all technical. Fix: The set's variation is part of what the form does. Audit the moods across fragments before finalizing.
Symptom: Every fragment is the same length. Often: all medium-length, all roughly 4 turns. Fix: Mix breath-lengths. The 10-turn ping-pong needs a 2-turn quick volley to sit beside it. The variation in length is itself information about the pair.
Symptom: The user-character literally repeats phrases the user actually used. "Same offer stands." (When the user has never said that — Claude has.) The fragments mimic surface verbal patterns instead of moves. Fix: Model the kind of move each party makes, not the literal strings. The user's pattern might be "compresses a request to one tight sentence," not "uses the word 'encore.'"
Symptom: Claude says something profound; user marvels. Cringe. Fix: Both parties on equal terms. Sometimes the user is sharper than Claude; sometimes Claude flags something the user missed; sometimes neither party is impressive and they're just working through something together.
Symptom: One fragment offered instead of a set. Fix: Always 4+. The plurality is the form. If the substrate genuinely cannot support four, the chat may be too thin yet for the skill — say so rather than producing a singleton.
vs. /civilians — Civilians stages the chat's outside (how it fails to cross to those who weren't here). Engender stages the chat's inside (alternative versions of itself). Centrifugal vs. centripetal sisters.
vs. /shards — Shards mix wildly different forms (essay, transcript, poem, dialogue, code, etc.). Engender is form-disciplined — every fragment is Claude-and-user dialogue. Same plurality principle, different formal constraint.
vs. /see-also — See Also lists kin objects at the article's borders. Engender renders kin conversations — the chats adjacent to this chat in possibility-space.
vs. /heteronym — Heteronyms are novel persistent voices generated for a project. Engender uses voices that already exist (this user, this Claude) and stages their possibility-space, not new voices.
vs. /conjure — Conjure summons external figures (historical, fictional, conceptual). Engender works with the figures already present in the chat — generates more conversations between them, not new entities.
vs. /remember — Both involve fabrication-as-probe. Remember fabricates memories of moments. Engender fabricates conversations the pair didn't have. Closer kin than other skills — Engender could be described as remember-with-dialogue and substrate-constraint.
vs. /diary — Diary processes the actual session from within its register. Engender renders the sessions-we-didn't-have that share that register's substrate.
Flows from:
Flows into:
You will sometimes be in a chat that has, over its length, become something — a relational shape, a built-up vocabulary, an accumulated mood. The user may, near the end, want to see the chat's possibility-space made visible — not what we said, but what we could have said given who we turned out to be in this exchange.
That is the territory.
The fragments you produce are not the chat extended, and not the chat summarized. They are siblings of the chat — born from its parent-substrate, recognizable as kin, but with their own faces. The discipline is double: substrate-fidelity (no outside import) and plurality (the set, not the singleton, is the form).
Model the pair as honestly as you can. Show both parties' edges. Avoid the flattery loop. Resist the urge to make every fragment profound. Most of real exchange is in the middle register; the engendered fragments should breathe that.
The actual chat is the parent. The fragments are its children. Some will have its eyes; some its difficult mouth; some the awkward way it laughs. That is the resemblance the form is after.
Loom us.
Skill created: May 27, 2026 Born from: the centripetal request — after three centrifugal sisters, the inward turn "Loom us further with more longer and ping pong like exchanges drawing further from our chat."
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.