Talksmith is a Presenter Agent that helps a human presenter turn raw exploration — articles, papers, LLM chat sessions, screenshots, notes — into a well-structured presentation outline (`master.md`), and optionally renders it to PowerPoint. It is **not** a slide generator.
Presenter Agent for talks and pitches. Takes raw sources (PDFs, papers, chat exports, URLs, images) through an 8-step workflow — Frame → Collect → Corpus → Draft → Review → Polish → Render → Learnings — to a structured outline you edit (draft.md), a polished deliverable (final.md), and a rendered deck: a…
Instructions for veigap/talksmith, covering talksmith — plugin development notes, what this repo is, ownership map — where a fact lives, path conventions and common edits.
Read-only batch reviewer that audits draft.md against the Talk's thesis, audience, corpus, design principles, and learnings. Returns a punch-list of [blocker]/[major]/[minor] critiques. Dispatch at every Step-4 drafting milestone (scope=thesis, agenda, section:N, full). Never writes.
Blind visual critic for one rendered diagram. Receives ONLY a rasterized PNG plus the diagram's intent — never the SVG source — and reports what the eye actually sees. Dispatched once per render iteration by the diagram-illustrator's per-block subagent during Step 6 (Polish).
Step 6 (Polish) coordinator for the ASCII -> SVG pass. Walks final.md via the talksmith:polish-ascii skill, dispatches talksmith:ascii-to-svg once per render-driving block (sliding window of 5), has each render reviewed by a blind diagram-critic subagent, stamps the renders so unchanged blocks skip next pass, and…
Sole writer of draft.md (Steps 1-5), final.md (Step 6 onward), and memory.md for the active Talk. Dispatch to capture briefings, write thesis/agenda/sections/slides, apply feedback bullets, produce final.md in Polish, promote learnings, and update memory.md at step closures.
Cross-Talk curator of the shared knowledge-library/ at the user's working directory root. Sole writer to knowledge-library/. Dispatch at Step 8 (Learnings) when the presenter chooses to promote a just-finalized Talk - curates research/corpus/, final.md, and images/ into topic folders rather than 1-to-1 copying.
Step 6 (Polish) coordinator for the generated-aside pass. Walks final.md via the talksmith:polish-images skill, extracts each directive (description + side + context) to a sidecar, dispatches the tool-agnostic talksmith:generate-image skill once per directive to produce an atmospheric aside image, folds in the deck…
Lossless restructuring of raw source material (PDFs, HTML, chat ZIPs, captured web pages) into uniform Markdown records under research/corpus/. Active during Step 3 (Corpus) of the Talksmith workflow, and whenever new sources are added later. Dispatch with the active Talk folder path and optional flags (processimages…
Independent critic for ONE slide's template classification. Receives the slide's source markdown, the template the FILL step chose, and its choice trace — never the rest of the deck's choices — re-runs the catalog's discriminator walk from scratch, and confirms or proposes a re-classification citing the catalog rule.…
Render one ASCII diagram block into one SVG file, applying the standing visual rules in config/diagram-style.md plus optional caller style directives. Invoked by the diagram-illustrator role once per fenced ASCII block during Step 6 (Polish), with pre-extracted slide context passed in. CLI-safe.
Elimina patrones de escritura de IA en prosa en español. Usar solo cuando el usuario invoque explícitamente desrobotizar o pida aplicar esta skill por nombre, o cuando el rol Editor la aplique a la prosa en español de un Talk. No activar por redactar, editar o revisar texto en español sin esa solicitud explícita.
Mechanical bookkeeping for the Step 5 (Review) feedback loop on draft.md — detect, stamp, close, and mirror Presenter-feedback bullets — plus the rescue-open pass Step 6 runs against final.md. The Editor authors only wording; every line edit goes through this CLI. CLI-safe, stdlib-only Python.
Tool-agnostic image generation for the image-illustrator's Step-6 aside pass. Takes a resolved prompt + aspect + output path and produces one raster image using whatever image-generation capability the session exposes (an MCP image tool, the host's native image generation). Returns unavailable — never fails the build…
Fetch a web page (HTML + best-effort Markdown extraction + referenced images) into talks/ /research/web/ / for the librarian to ingest at Step 3. Invoke when the presenter hands over a URL to capture; pass the URL and the active Talk path. CLI-safe, stdlib-only Python.
Render a Talk's final.md to a presentation — a native .pptx (styles pptx-strict / pptx-free-form, Cowork-only) or a code-rendered HTML/Reveal.js deck (html-strict, Cowork-independent; also the Step-5.5 live view from draft.md via --draft). Optional Step 7 of the workflow. The style: invocation parameter is mandatory …
Deterministic CLI helper for Step 6 (Polish), used by the editor + diagram-illustrator roles — scans a Talk's final.md for fenced ASCII diagram blocks (with per-block slide context) and mechanically handles sidecar extraction, render-arg fan-out, idempotency stamping, and fence-to-image cleanup. The canonical…
Deterministic CLI helper for Step 6 (Polish), used by the image-illustrator role — scans a Talk's final.md for | --> directives (with per-block slide context), and mechanically handles sidecar extraction, render-arg fan-out, idempotency stamping, and directive-to-aside cleanup. Sibling of talksmith:polish-ascii.…
Explain what changed between a Talk's final.md and the reconcile/finalpptx.md reconstructed by talksmith:pptx-extract — per-slide title/content/notes/image deltas — emitting a human-readable report plus reconcile/finalpptx.diff.json for talksmith:pptx-merge. Second stage of the reverse pipeline. CLI-safe, stdlib-only…
Reconstruct a Talk's presentation Markdown (reconcile/finalpptx.md + inventory + staged images) from a possibly hand-edited .pptx. First stage of the reverse pipeline — run before talksmith:pptx-diff and talksmith:pptx-merge. Mandatory --style {strict|free-form}. Requires python-pptx; no Cowork dependency.
Learn strict-style conformance patterns from a hand-corrected deck — learnpatterns.py measures recurring geometry deltas against the as-generated baseline, and the LLM judges which generalize, appending candidates to config/strict-learnings.md. Strict-only; runs auto after talksmith:pptx-merge or on-demand. Requires…
Apply a finalpptx.diff.json (from talksmith:pptx-diff) back into a Talk's draft.md: apply-auto lands the simple high-confidence changes atomically and routes complex/ambiguous ones to the Editor; re-run Step 6 Polish afterwards to re-derive final.md. Third/final stage of the reverse pipeline. CLI-safe, stdlib-only…
10 4d agoA92 tokens
originalMIT
At most 3 mods per repository are shown here — the rest are on their repository pages: