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 commands/josephscript/claude-speak/speakgit clone --depth 1 https://github.com/JosephScript/claude-speakWhat 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 | $0.00026 | $0.00393 |
| Opus 5 | $0.00013 | $0.00197 |
| Sonnet 5 | $0.00005 | $0.00079 |
| Haiku 4.5 | $0.00003 | $0.00039 |
Grade B, and why
speak scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
2. Otherwise, read the persisted default: run `cat ~/.claude/claude-speak/mode 2>/dev/null` via Bash. If it outputs exactly `llm`, use llm mode. Otherwise (empty, missing file, or anything else), use local mode. Toggle t What it actually says
Determine which mode to use for this call:
- If the argument passed to this command is exactly
llmor exactlylocal, use that mode for this call only. - Otherwise, read the persisted default: run
cat ~/.claude/claude-speak/mode 2>/dev/nullvia Bash. If it outputs exactlyllm, use llm mode. Otherwise (empty, missing file, or anything else), use local mode. Toggle the persisted default with/toggle-speak-mode.
Local mode (mechanical, the default): run this exact command via Bash and do nothing else besides a brief confirmation once it finishes:
"${CLAUDE_PLUGIN_ROOT}/scripts/speak-last-response.sh"
This script reads your last response straight from the session transcript, strips markdown, truncates at a sentence boundary if long, and speaks it verbatim. Do not read, paraphrase, or summarize the response yourself in this mode — just run the script.
LLM mode: take the plain-text content of your own response immediately preceding this command (not this instruction, not older turns) and speak it aloud by piping it into "${CLAUDE_PLUGIN_ROOT}/scripts/speak-say.sh" via Bash, e.g. printf '%s' "<text>" | "${CLAUDE_PLUGIN_ROOT}/scripts/speak-say.sh" (this applies the persisted rate/voice; don't call the speech engine directly here). If that response is longer than roughly 600 characters, don't read it verbatim — write a concise, spoken-friendly summary of it instead.
Don't narrate which mode you picked, or that you're doing this, beyond a brief confirmation once speech has finished.
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.
- 2d ago First seen · 21 lines · 26 tokens per session scan B a53f61eee246
speak is a command published in the GitHub repository JosephScript/claude-speak (3 stars, last pushed 29d ago), licensed MIT. It adds 26 tokens to every session and 393 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.