Tender Document Review Checklist

Problem

Tender review often gets stuck because clauses are long, risks are scattered across the main text, attachments, and scoring rules, and pre-bid checks are hard to verify line by line. This skill serves bidders and does not decide whether to bid or suggest pricing; it outputs fact-based checklists with evidence anchors to catch disqualification items, scoring items, marked items such as ▲/★/※, and contract-stage constraints.

How it works

• Read references/workflow-sop.md and references/output-schema.md, then follow a fixed workflow.
• Use init_run.py to create output directories, then extract .lines.txt and .tables.json; original line numbers are the audit anchors.
• Run scan_keywords.py and scan_candidates.py for broad scanning, then use the model to judge whether hits are real bid, evaluation, or contract risks.
• AI-discovered words must first be rescanned in the current tender; harvest_ai_words.py creates a review list, and only user approval writes them to the local keyword store.
• Do not generate HTML or Excel until quality_gate.py passes, and run red-team review and cross-checks so the output is traceable and schema-compliant.

Scope

Best for pre-bid self-check. It does not provide legal advice, decide bid/no-bid, set pricing, or rank business priorities. If quality_gate.py or check_worklist_schema.py is missing, the skill stops and reports that the package is incomplete.