Files

altair823 2c058ab175 feat(rag): multi-turn ask — Turn struct + ask_with_history + token budget (p9-fb-15)

Spec PR #59 의 §3.8 multi-turn behaviour 구현. RAG facade 가 prior
turns 받아 prompt 에 prepend, retrieval query expansion 적용,
Answer 에 conversation_id / turn_index 채움.

신규 (kebab-core):
- Answer 에 conversation_id (Option<String>) / turn_index (Option<u32>)
  field 추가. serde skip_serializing_if 로 single-shot 의 wire
  output 변경 0 (기존 외부 wrapper 영향 없음).
- Turn struct (question + answer + citations + created_at).
- RefusalReason::LlmStreamAborted variant.

신규 (kebab-rag):
- AskOpts 에 history (Vec<Turn>) / conversation_id / turn_index 3 field.
- AskOpts::single_shot(mode) helper.
- RagPipeline::ask_with_history(query, history, conversation_id,
  turn_index, opts) — combined opts 로 ask 호출.
- expand_query_with_history: history.last() 의 answer 첫 200 자
  concat 해 SearchQuery.text 확장 (spec §3.8 의 \"cheap concat\";
  LLM-based standalone-question rewriting 은 P+).
- serialize_history + remaining_history_budget_chars: spec 의 priority
  enforcement — system+question 필수, retrieved chunks 가 차지한
  뒤 남은 char budget 안에서 newest 우선, oldest drop.
- ask 본문: history 가 비어있지 않으면 [이전 대화] 블록을 user
  prompt 위에 prepend. Answer 생성 site 3 곳 (정상 / NoChunks /
  ScoreGate refuse) 모두 conversation_id / turn_index 채움.

신규 (kebab-store-sqlite):
- refusal_reason_label 가 LlmStreamAborted → 'llm_stream_aborted'.

기존 caller 변경 (single-shot 동작 동일):
- kebab-cli main.rs Cmd::Ask: AskOpts 에 history=Vec::new(),
  conversation_id=None, turn_index=None 명시 (CLI multi-turn 은
  p9-fb-18 의 --session/--repl 가 채움).
- kebab-tui src/ask.rs spawn site 동일 (multi-turn UI 는 p9-fb-16).
- kebab-eval runner.rs golden eval 동일 (single-shot per query).
- kebab-app tests/ask_smoke.rs / kebab-tui tests/ask.rs / kebab-rag
  tests/pipeline.rs / kebab-eval metrics.rs Answer literal 갱신.

Test:
- 9 신규 lib unit (expand_query 4 / serialize_history 3 / remaining_budget 2).
- 기존 12 PASS 회귀 0.

Plan 갱신:
- p9-fb-15 status planned → in_progress. 머지 후 한 줄 commit
  으로 completed flip.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-02 23:09:46 +00:00

2.4 KiB

Raw Blame History

phase, component, task_id, title, status, depends_on, unblocks, contract_source, contract_sections, source_feedback

phase

component

task_id

title

status

depends_on

unblocks

contract_source

contract_sections

source_feedback

kebab-rag + kebab-app

p9-fb-15

RAG multi-turn — history-aware prompt + token budget

in_progress

p9-fb-16

p9-fb-17

p9-fb-18

../../docs/superpowers/specs/2026-04-27-kebab-final-form-design.md

§7 RAG

p9-dogfooding-feedback.md item 13

p9-fb-15 — RAG multi-turn core

Goal

kebab-rag 가 conversation history (Vec<Turn>) 를 받아 prompt 빌드. token budget 안에서 retrieval k 와 history truncation 정책으로 fit.

Allowed dependencies

기존 kebab-rag deps.
tiktoken-rs 또는 LLM family-specific tokenizer (gemma 토큰화). 우선 char 기반 ÷4 근사 (cheap & 의존 X).

Public surface

pub struct Turn {
    pub question: String,
    pub answer: String,
    pub citations: Vec<Citation>,
    pub ts: OffsetDateTime,
}

pub fn ask_with_history(
    cfg: &Config,
    new_question: &str,
    history: &[Turn],
    stream: Sender<RagEvent>,
) -> anyhow::Result<Answer>;

kebab-app 도 ask_with_config_and_history(cfg, q, history, stream) 추가.

Behavior contract

prompt 구조: system_prompt + history_serialized + retrieved_chunks + new_question. 형식 (roles 또는 plain text) 는 prompt_template_version bump (rag-v1 → rag-v2).
token budget: cfg.rag.max_context_tokens.
- 우선순위: system + new_question 항상 포함.
- 다음: retrieved chunks (k=cfg.search.default_k 부터, budget 초과시 k 감소).
- 마지막: history. budget 남은 만큼 newest turn 부터 포함, 부족하면 oldest turn drop. 최소 0 turn 까지 가능.
retrieval query: new_question + " " + last_turn.answer.first_N_chars(200) concat (cheap query expansion). LLM 기반 standalone question rewriting 은 P+.
streaming: RagEvent::Token(s) / RagEvent::Done(answer) / RagEvent::Error(e).

Test plan

kind	description
unit	history 5 turn → token budget 초과 시 oldest 부터 drop
unit	retrieved_chunks vs history 의 priority
integration	가짜 history (Q1/A1) + new Q2 → prompt 에 Q1/A1 포함 (snapshot)

DoD

cargo test -p kebab-rag -p kebab-app 통과
prompt_template_version bump (rag-v2)
HOTFIXES X (신규)
frozen design §7 RAG 절 갱신 (multi-turn 정책)

Out of scope

LLM 기반 question rewriting (P+)
conversation 영속화 (p9-fb-17)
UI (p9-fb-16, p9-fb-18)

2.4 KiB Raw Blame History