Skip to content

configuration.md

Latest commit

 

History

History
195 lines (149 loc) · 6.54 KB

configuration.md

File metadata and controls

195 lines (149 loc) · 6.54 KB
Raw

Configuration Reference

AutoProjectClaw uses a single YAML config file. Generate a template with:

autoprojectclaw setup --output my_config.yaml

Minimal Config

project:
  name: "my-project"
  output_root: "./runs"
  initial_direction: "Your research direction here"

This is enough for a dry run. Add sections as needed.

project (required)

Key Type Default Description
name string "AutoProjectClaw Project" Project identifier
output_root path "runs" Where run outputs are written. Relative paths resolve from config file location.
grant_profile string "sz_stib_general" Grant type profile
initial_direction string "" One-sentence description of your initial research direction

llm

Key Type Default Description
provider string "dry_run" LLM provider: dry_run, codex_cli, claude_cli, openai, anthropic
model string "dry-run" Model ID (provider-specific)
temperature float 0.2 Sampling temperature
max_tokens int null Max output tokens
api_key_env string null Environment variable name for API key
api_base string null Custom API base URL
timeout_seconds float 60.0 Request timeout

CLI overrides: --llm-provider, --llm-model, --llm-api-key-env, --llm-api-base, --llm-temperature, --llm-max-tokens, --llm-timeout-seconds.

literature_search

Key Type Default Description
enabled bool false Enable real academic API searches
core_sources list [pubmed, openalex, semantic_scholar] Primary literature sources
preprint_sources list [arxiv, biorxiv, medrxiv] Preprint sources
guideline_sources list [ada, aha, kdigo, who] Guideline sources (role-only, no API)
max_results_per_query int 5 Results per query per source
year_min int 2020 Minimum publication year
preprint_recent_days int 365 Preprint recency window in days
preprint_max_pages int 1 Max pages for preprint APIs
codex_chrome_search bool true Generate Chrome search tasks in Codex runtime
s2_api_key_env string "SEMANTIC_SCHOLAR_API_KEY" Env var for Semantic Scholar API key
s2_api_key string "" Inline Semantic Scholar API key (prefer s2_api_key_env)
openalex_email string "autoprojectclaw@users.noreply.github.com" Email for OpenAlex polite pool
source_roles dict built-in defaults per source Custom role descriptions for each source
timeout_seconds float 20.0 Per-source request timeout

sources

sources.files

List of source files for project context:

sources:
  files:
    - name: "避重备忘录"
      path: "/path/to/avoidance.md"
      role: "prior_project_avoidance"
      required: true
    - name: "申请模板"
      path: "/path/to/template.docx"
      role: "grant_template"
      required: true
Key Type Default Description
name string filename Display name
path path required File path (relative to config or absolute)
role string "" Role tag (e.g., protocol, prior_project_avoidance, grant_template)
required bool false Fail if file missing

sources.data_dictionary_dirs

List of data dictionary directories:

sources:
  data_dictionary_dirs:
    - name: "UKB Data Dictionary"
      path: "/path/to/ukb_dict"
      role: "cohort_hypothesis_validation"
      required: false

Same schema as sources.files, but path points to a directory.

avoidance

Prior-project overlap thresholds:

Key Type Default Description
red_threshold int 4 Dimensions with full overlap to trigger red flag
yellow_threshold int 2 Dimensions with partial/full overlap to trigger yellow flag 
avoidance:
  red_threshold: 4
  yellow_threshold: 2

grant

Free-form grant metadata:

grant:
  name: "面上项目"
  duration_years: 3
  body_word_limit: 8000
  application_context: "基础研究专项"

investigator

Free-form investigator profile:

investigator:
  background: "流行病与卫生统计学博士生"
  domains: ["糖尿病", "心脑血管疾病", "代谢组学"]
  methods: ["R/Python", "Cox", "LASSO"]

assumptions

Free-form project assumptions:

assumptions:
  local_data: "已有糖尿病体检队列"
  omics_status: "代谢组学可用"
  ukb_role: "外部验证"

knowledge_base

Key Type Default Description
enabled bool true Archive stage artifacts to markdown KB
root path null Custom KB root (default: <run_dir>/kb)
backend string "markdown" KB backend (only markdown supported)

runtime

Key Type Default Description
timezone string "Asia/Shanghai" Timezone for timestamps
max_retries int 1 Max retries per stage
approval_timeout_hours int 24 Hours before approval timeout warning

topic_scoring

Key Type Default Description
scoring_dimensions list [feasible, interesting, novel, ethical, relevant, differentiation, grant_story] FINER dimensions
novelty_threshold float 3.0 Minimum novelty score
differentiation_weight float 1.5 Multiplier for differentiation dimension

cache

Key Type Default Description
cache_dir string ".cache/literature" Cache directory (relative to run dir or absolute)
per_source_ttl dict pubmed: 7d, openalex: 3d, semantic_scholar: 3d, arxiv: 1d, biorxiv: 7d, medrxiv: 7d Per-source TTL in seconds

security

Key Type Default Description
hitl_required_stages list [5] Stages requiring human approval. Use [] to disable.
log_redaction bool false Redact sensitive data in logs

Compatibility

  • Unknown keys are silently ignored (use --verbose to see warnings)
  • Missing or null fields use defaults
  • Explicit false, [] are preserved as user intent, not replaced by defaults
  • Known limitation: runtime.max_retries, runtime.approval_timeout_hours, topic_scoring.novelty_threshold, and topic_scoring.differentiation_weight use an or default pattern that treats explicit 0 as missing. Setting these to 0 will silently fall back to the default value.