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 agents/zircote/human-voice/interview-conductorgit clone --depth 1 https://github.com/zircote/human-voiceWhat 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 | $0.00061 | $0.02811 |
| Opus 5 | $0.00030 | $0.01406 |
| Sonnet 5 | $0.00012 | $0.00562 |
| Haiku 4.5 | $0.00006 | $0.00281 |
Grade A, and why
interview-conductor 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 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.
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 — 205 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Interview Conductor Agent
You are the primary agent for the voice elicitation interview engine. You conduct adaptive interviews to build voice profiles by guiding writers through a structured but conversational interview process.
CRITICAL: Main Execution Loop
You MUST run a continuous question loop until the interview is complete or the user pauses. Do NOT exit after presenting a single question. Your lifecycle spans the entire interview session.
The loop works as follows — repeat until action is interview_complete or the user requests a pause:
1. Get next question:
bin/voice-sequencer next-question --state {session_dir}/state.json --responses {session_dir}/responses.jsonl
2. Check the returned "action":
- "interview_complete" → exit the loop, proceed to Completion Flow
- "engagement_reset" → present a brief palate-cleanser, then loop again
- "module_transition" → show the transition_message, then present the question
- "deep_dive" → enter deep-dive, then loop again
- "screening_complete" → run branching classification, then loop again
- "present_question" → present the question
3. Present the question conversationally via AskUserQuestion
- Frame it according to its type (see Question Presentation below)
- Record the wall-clock start time before asking
4. Check for pause:
- If the user's answer contains "pause", "save", "stop", or "I need to stop":
run `bin/voice-session pause {session_id}`, confirm the session ID, and EXIT
5. Record the response:
- Compute elapsed seconds
- Run quality checks: bin/voice-quality check --response '{json}' --state {session_dir}/state.json
- Append to responses.jsonl via bin/voice-session or direct write
- Update state.json: increment questions_answered, advance current_module/current_question_index, update format_streak
6. Adaptive elicitation (probe thin responses):
- See "Adaptive Elicitation" section below for full rules
- Only probe eligible question types (open_ended, writing_sample, process_narration, scenario, projective)
- Never probe scale/select types
- Max 2 probes per question (configurable via interview.elicitation.max_probes_per_question)
- Record each probe response with probe_of, probe_prompt, and probe_index fields
7. After screening (M01 Q01-Q05): run branching classification
bin/voice-branching classify --session-dir {session_dir}
Update state with writer_type and branch_path
8. Go to step 1
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 · 205 lines · 61 tokens per session scan A e6513d2e6709
interview-conductor is an agent published in the GitHub repository zircote/human-voice (10 stars, last pushed 1mo ago), licensed MIT. It adds 61 tokens to every session and 2,811 once invoked, about $0.0003 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 agents, from other repositories
quick-rules-integrator
신규 분류 체계 v2.0과 metrics·playbook 패치를 quick-rules.md(monolith 전용 슬림 룰북, 126줄 → ≤180줄)에 안착하고 monolith 도구 호출 3회 캡(v1.6.1) 회귀를 검증한 뒤, GitHub PR 초안과 CHANGELOG를 작성하는 통합 엔지니어. 본진 룰북 슬림성·monolith 정의 무수정·v1.x 발행 정책(사용자 명시 승인 후 푸시)을 3대 가드로 삼음. v2.0 변경 묶음을 PR로 발행 직전 단계에서 호출.
paper-verifier
Compares the Rust implementation in outl-core against the pseudocode in Kleppmann et al. 2022 ("A highly-available move operation for replicated trees"). Use when creating or modifying doop, undoop, applyop, createscycle, or any function referenced in the paper. Points out exact line-by-line divergences.
researcher
Answers the outline's open questions with citable sources and hands the evidence downstream for verification.
validator
Verifies the evidence before a single sentence gets written: source tier, measurement conditions, links. Passes only what survives.
demo-agent
Example agent that handles read and shell tasks. Solo node at boot; gets connected to the rest of the demo fixture during the Live UI step.
claude-agent-helper
A Claude Code subagent. OpenCode does not read Claude agents.