test-before-bulklisted

# Test Before Bulk Convention Lifted from a shared brain-conventions repo as a convention applied across CRM, knowledge-graph, spreadsheet, vector-index, and any other durable-write skill. **Never run a batch operation without testing one first.** ## Tone & Voice Adversarial convention. applies as a gate, not a suggestion. **NON-NEGOTIABLE**: if the user says "just run it on all 170" without a test-on-1 step, push back. The test-first discipline exists because durable writes (CRM rows, knowledge-graph pages, vector embeddings) compound errors at scale and are hard to unwind. One failed bulk = hours of cleanup. ## Output format The skill enforces a procedural gate, not an artifact. The outcome is: 1. Confirmed-good output from test-on-1 (artifact pinned + diffed against the expected format) 2. Updated skill code/prompt (if the test revealed a bug. fix the SKILL, not a one-off script) 3. Bulk run with throttling + kill switch (e.g. `--limit N` or `--dry-run` first) 4. Post-bulk: spot-check 3 random items from the bulk output to verify the pattern held ## The Process (6 steps) 1. **Read the existing skill first.** Don't write throwaway scripts. If a skill exists for the operation (e.g. a CRM operator, an ingestion processor, a contact extractor), use it. If a skill doesn't exist and the bulk op will be repeated, the right move is to BUILD a skill, not a one-off bash. 2. **Hone the prompt/logic on a single item.** Get the output format right before running anything. If th