Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/veigap/talksmith/claude-mdgit clone --depth 1 https://github.com/veigap/talksmithWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.06941 | $0.06941 |
| Opus 5 | $0.03470 | $0.03470 |
| Sonnet 5 | $0.01388 | $0.01388 |
| Haiku 4.5 | $0.00694 | $0.00694 |
Grade A, and why
talksmith CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 161 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Talksmith — plugin development notes
This file is the project-instruction file for plugin authors and contributors working in this repo. It is loaded by Claude Code only when a session is opened at the root of this repo to develop the plugin. End users never see it.
Two unrelated
CLAUDE.mdfiles — don't confuse them.
- This file (
/Users/.../talksmith/CLAUDE.md) is the plugin source repo's dev notes. It exists only here.- A user's
CLAUDE.mdis a per-directory stub that activates Talksmith for one subject working directory. It is created by/talksmith:initfromtalksmith-orch.md, lives in the user's cwd, and is completely separate from this file.Installing the plugin (
/plugin install talksmith@talksmith) is a one-time, machine-wide action — it does not create anyCLAUDE.mdanywhere. Initializing Talksmith for a working directory (/talksmith:init) is a separate, per-directory action that writes the stub. A user can install the plugin once and then run/talksmith:initin many different directories.
For the user-facing project overview, see README.md. For the full Presenter Agent operating spec (eight subagents, eight steps, schemas, interaction defaults), see orchestrator.md — that file stays in the plugin install and is auto-imported at session start by the thin talksmith-orch.md stub that /talksmith:init writes into a user's subject working directory.
What this repo is
The Talksmith Claude Code plugin. Installable surface:
| Path | Purpose |
|---|---|
.claude-plugin/plugin.json |
Plugin manifest (name, version, description). |
agents/ |
Eight Claude Code subagents — librarian, composer, editor, diagram-illustrator, image-illustrator, diagram-critic, slide-classifier-critic, global-librarian. Each has YAML frontmatter (name:, description:) so it can be dispatched by name. diagram-illustrator and image-illustrator are Step-6 siblings: the former renders authored ASCII → SVG (diagrams); the latter generates atmospheric aside imagery from <!-- generate-image: … --> directives (tool-agnostic, degrades when no image capability is present). diagram-critic is the odd one out: it is dispatched by the diagram-illustrator's per-block subagent rather than by the orchestrator, and its tools: Read restriction is load-bearing — it reviews a rendered diagram from the PNG alone, and never receives the SVG path. That blindness is the point (see its own file). slide-classifier-critic is its twin one artifact earlier: dispatched by the md-to-deck skill (not the orchestrator) once per content slide between FILL and RENDER, it re-runs the catalog's discriminator walk on one slide's source and confirms or overturns the template the fill chose. Its tools: Read restriction is load-bearing for the same reason — it is blind to every other slide's classification, because a critic that can see the deck is concept-breakdown twenty times reads the twenty-first as normal. |
commands/ |
Slash commands. Currently one: /talksmith:init. |
skills/ |
Thirteen skills, all namespaced talksmith:<skill> in their SKILL.md frontmatter (the namespace is the convention — an un-namespaced name: collides with a user's own same-named skill instead of coexisting with it): ingest (web capture), ascii-to-svg (one ASCII block → one SVG; owns the viewBox, rasterizer, and aspect-audit contracts), polish-ascii (Step-6 mechanical scan / sidecar / stamp / cleanup for diagrams — stamp-renders arms render idempotency), polish-images (its sibling — the same staged scan / sidecar / stamp / cleanup for <!-- generate-image: … --> aside directives) and generate-image (tool-agnostic backend the image-illustrator dispatches; produces a raster or returns unavailable, never blocks), feedback-cycle (Step-5 bookkeeping CLI), md-to-deck (Step-7 render, three modes), and the reverse pipeline pptx-extract → pptx-diff → pptx-merge (reconcile an externally-edited .pptx back into draft.md; artifacts under talks/<Talk>/reconcile/) plus pptx-learn (mines hand-corrections into strict conformance-pattern candidates). Shared code lives once under skills/_shared/, sys.path-imported by its consumers: _pptxlib.py (pptx parsing — the three reverse-pipeline scripts), _context.py (the slide-context scanner — both polish skills), _plan.py (the plan-file envelope, --final/--plan/--dry-run, and the gc live set — both polish skills, which are otherwise near-twins), and _write.py (the atomic in-place write every skill that edits draft.md/final.md must use — feedback-cycle + pptx-merge). The deck-parsing audits have their own sibling module, skills/md-to-deck/audits/_ooxml.py (namespace map, slide relationships, solid-fill colour, PNG dimensions, and the _-key-excluding model walk). Two more are anti-slop passes the Editor applies to a Talk's prose, one per language: stop-slop (English, vendored from Hardik Pandya) and desrobotizar (Spanish) — both are explicit-invocation only, so neither fires on ordinary drafting or editing; the dispatch table lives in agents/editor.md. pptx-extract / pptx-learn need python-pptx; none require Cowork. Everything else lives in each SKILL.md — don't restate it here. |
schemas/ |
File-format specs (canonical empty forms for draft.md, memory.md, profile.md, principles.md, learnings.md, feedback-backlog.md, feedback-processed.md, corpus-record.md, talksmith-bugs.md), plus slide-model.md — the structured slide-model.json the md-to-deck skill fills (LLM) and the renderers consume. |
config/principles.md, config/diagram-style.md, config/pptx-styles/ |
Bundled read-only design assets. |
talksmith-orch.md |
The thin working-directory stub copied to CLAUDE.md in the user's cwd by /talksmith:init. Tells the agent to load the full spec from ${CLAUDE_PLUGIN_ROOT}/orchestrator.md at session start. Edit only when the session-start contract changes (new mandatory load, new directive) — and warn users to re-run /talksmith:init when you do. |
talksmith-agents.md |
The Codex working-directory stub copied to AGENTS.md in the user's cwd by /talksmith:init. A thin pointer to CLAUDE.md (which owns the boot instructions) that re-states the two Claude-Code-specific fallbacks a non-Claude-Code agent needs — the @-import is inert, ${CLAUDE_PLUGIN_ROOT} is unset. Carries no independent workflow content, so it never drifts from CLAUDE.md. Edit only when those fallbacks change; re-init to redeploy. |
orchestrator.md |
The full Presenter Agent operating spec — workflow Steps 0 → 8, role dispatch, schemas, interaction defaults. Stays in the plugin install; never copied. Edit freely when you change workflow, role contracts, or step-by-step behavior — plugin updates roll out automatically and existing user working directories pick up the change on their next session reload, without re-init. |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 161 lines · 6,941 tokens per session scan A d4ff43587f9f
talksmith CLAUDE.md is an instructions file published in the GitHub repository veigap/talksmith (10 stars, last pushed 5d ago), licensed MIT. It adds 6,941 tokens to every session, about $0.0347 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
spec-kit AGENTS.md
AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.
codex AGENTS.md
AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).
next.js AGENTS.md
Instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.