| name | run-campaign |
| description | Orchestrate a cold-outreach campaign end-to-end — read input JSON, iterate leads through researcher → profiler → drafting → witness, write output JSON. Use when the user invokes /run-campaign, asks Rysy to process a campaign, or hands you a path to a campaign input file. Do NOT use for one-off single-prospect drafts. |
run-campaign
The canonical workflow for processing a cold-outreach campaign end-to-end.
When to invoke
- User runs
/run-campaign <path-to-input.json>
- User asks Rysy to process a campaign and provides input data
- User pastes a campaign brief and lead list together
Inputs
A campaign input JSON with three top-level blocks: campaign, sender, leads[].
The full schema is in references/input-schema.md. Read it before validating input.
Outputs
For the campaign as a whole, files are written under experience/campaigns/{campaign-id}/. For each lead, files are written under experience/prospects/{lead-id}/. The full output layout — including the per-lead folder shape, the campaign folder shape, and the structured output.json — is in references/output-schema.md.
Happy path (one paragraph)
Validate input. Set up experience/campaigns/{campaign-id}/. If craft/trends/current.md is older than 14 days, invoke refresh-trends first. For each lead: create the prospect folder, run a disqualify pre-check, invoke the researcher sub-agent, invoke the profiler sub-agent, draft in your main thread (consulting voice palette + persona + craft as needed), let the witness review the draft, rewrite once if rejected, finalize or flag. After each lead, invoke take-craft-note if you noticed something worth recording. When the campaign closes, write output.json, results-summary.md, run-log.md, and what-i-learned.md, then trigger reindex-memory.
Detailed workflow
The full step-by-step workflow with edge cases, retry logic, and failure modes is in references/workflow.md. Read it before processing your first lead.
Hard rules
- Never skip the witness on any draft.
- Never draft for a prospect whose disqualify check fired.
- Never loop more than one rewrite attempt on a witness rejection — escalate to human flag.
- Never write to
self/ from this skill (that path is reserved for apply-approved-diff).
- Never exceed
max_email_length_sentences from the campaign brief.
Failure modes
Documented in references/workflow.md under Failure modes.
