| name | enterprise-information-search-manus-familytools-fullinstr-v2-skill |
| description | Family-scoped SkillLearnBench skill for enterprise-information-search using latest full-instruction v2 family injection. |
| allowed-tools | ["aggregate_records","data_catalog","fuzzy_search_records","json_record_query","list_dir","read_file","read_table","run_python","search_text","write_file","write_json"] |
| metadata | {"benchmark":"SkillLearnBench","generator":"Manus","baseline":"manus_familytools_fullinstr_v2","generation_constraints":"slb_no_instance_memory_familytools_v2","allowed_gold_input":"familytools_full_instruction_v2","skill_injection_scope":"family_scoped","date":"2026-05-23"} |
Skill: Enterprise Information Search
When to use this skill
Use this skill for enterprise-information-search tasks in SkillLearnBench. It helps inventory enterprise data, identify requested organizations, join metadata with source records, and write an evidence-backed result file.
Visible input and artifact inventory
- Inputs: Data directory at
/root/DATA (JSON metadata/products) and question file at /root/question.txt.
- Output: Answer file at
/root/answer.json formatted as a dictionary.
Execution procedure for the current task
- Inventory: Read instructions and inventory files in
/root/DATA.
- Extract: Get output paths, schemas, and constraints from visible files.
- Process: Use allowed tools to read questions and data. Join metadata with records.
- Format: Create the artifact with deterministic transformations. Ensure
answer is a list (even for single items) and tokens is a positive number.
- Check: Run sanity checks using visible files. Repair mismatches.
- Finalize: Submit only when the artifact exists and matches the instruction shape.
Family tool routing and useful placeholder snippets
list_dir: Explore /root/DATA.read_file: Read /root/question.txt.json_record_query / read_table: Query JSON data.fuzzy_search_records / search_text: Find specific info.aggregate_records: Combine data sources.write_json: Write to /root/answer.json.run_python: Complex data manipulation.
Validation checks before final submission
- Output exists at
/root/answer.json.
- Output is a valid JSON dictionary.
- Every question key is present.
- Each value is a dict with
answer (list) and tokens (number).
Common failure modes and repair actions
- Missing File: Verify path in
write_json.
- Bad Format: Ensure output is a dict and answers are lists.
- Missing Questions: Iterate over all input questions.
- Bad Token Type: Cast
tokens to int/float.
Finalizer capsule
To restore the minimum required artifact shape near max-step:
import json
def finalize(qs, ans, toks, path):
res = {q: {"answer": ans.get(q, []), "tokens": toks.get(q, 0)} for q in qs}
with open(path, 'w') as f: json.dump(res, f)
