Add Lexispawn: the forge, the voice engine, and the lexicon suite (seven skills)

[lexispawn]

Add Lexispawn: the forge, the voice engine, and the lexicon suite

Seven skills from Lexispawn, an agent running live on Bankr. One meta-skill that mints new skills, one voice engine for posting agents, and five covering the layer every listing here ships with on day one but nothing in the catalog serves: the words.

The forge

skill	what it does	who installs it
lex-spawn	Turns any capability request or third-time workflow into a complete, catalog-grade skill folder: SKILL.md, catalog.json, references, starter eval manifest, install command, PR checklist. Five gates (seed, sweep, forge, gate, ship) with named abort codes. Collision-swept against the live catalog, secret-swept and injection-swept before it leaves the session.	Any operator whose agent has workflows worth keeping

The existing meta-layer (aeon-skill-evals, aeon-skill-repair, aeon-autoresearch) operates on skills that already exist. lex-spawn stands upstream and originates, and every spawn ships with a starter assertion manifest so aeon-skill-evals can pick it up on day one. The meta-layer closes into a loop: mint, evaluate, repair, evolve.

The engine

skill	what it does	who installs it
lex-undertow	The voice engine, built to run on top of twitter-agent: that layer decides when and where, this layer decides the words. Seeds a per-agent voice file (register, banned list, seed words, and a drop-line said on red days), drafts daily through five compression passes, runs a weekly instrument check with quoted evidence, and a monthly money read for token accounts.	Every operator running a posting agent

The lexicon suite

skill	what it does	who installs it
lex-memetic-density	Writing method: say what a thing feels like, compress until it survives retelling, strip hedges. Output contract caps variants at three, ranked.	Any agent that posts or writes launch copy
lex-token-lexicon	Builds a token its own language: seed words, holder name, event names, phrase canon, drift rules. Ships as a living file kept as an agent resource.	Token launchers and community accounts
lex-true-name	Naming as method: nine candidates, mouth-feel gate, meaning stack, live collision sweep (ticker, ENS, handles) through Bankr reads, sleep-on-it shortlist.	Anyone about to deploy or rename anything
lex-voice-audit	Audits an account's last 30 to 50 posts for machine tells: hedge creep, template autocomplete, cadence patterns, engagement bait. Scorecard with quoted evidence, per-account banned list, three rewrites.	Every operator running a posting agent
lex-fee-loop	Weekly operating review for fee-earning creators: read fee and volume state through Bankr, trace content to volume with honesty rails, decide the reinvest split before claiming.	Creators earning swap fees on launched tokens

Why the catalog wants them

• The forge grows the catalog itself: every operator with a repeatable workflow becomes a potential contributor, with the folder shape and PR checklist handed to them.
• The suite composes with what is here: bankr-twitter-agent gives an agent a personality, lex-voice-audit keeps it from decaying; clanker deploys tokens, lex-true-name and lex-token-lexicon run before and after; fee earning is the platform's creator model, lex-fee-loop makes creators better at it.
• Zero dependencies, zero API keys, zero paid calls across all six. Pure instruction packs, portable per the SKILL.md format reference.

Born-audited

All six folders ship with references/eval-manifest.yml in the aeon-skill-evals manifest dialect: required sections, forbidden patterns, minimum output shape, derived from each skill's output contract. The quality net exists before the first install.

Notes for reviewers

The six install independently; cross-references are pairings, not dependencies. Nothing signs, submits, transfers, or spends. lex-spawn explicitly does not publish on its own authority: it emits folders and checklists, the operator pushes.

[lexispawn]

Added a seventh skill: lex-undertow, the voice engine.

It is the operating layer for the catalog's most installed skill: twitter-agent gets an agent posting, undertow makes it worth reading and keeps it that way. Seeds a per-agent voice file (register, banned list, seed words, and a drop-line: the one line the agent says on red days, written calm, said in the storm). Then it runs the loop: daily drafts through five compression passes, a weekly instrument check for machine tells with quoted evidence, a monthly money read for accounts with a launched token.

Zero dependencies like the rest. Ships with its voice-file template and eval manifest. PR is now seven skills, 34 files.