Update to august

{
  "name": "feedback-loops",
  "description": "Run an iterative feedback loop with August twins: put the same artifact (copy, design, concept, variant set) in front of a panel of twins round after round, applying the top suggestion from each round as a new variant, until the responses converge on a preferred direction. Auto-invokes /fix-question when a twin response is vague or hedged and re-asks that twin with a sharper prompt. Meta-skill that orchestrates other August skills so you don't have to manually kick each round. Triggers on: \"run a feedback loop\", \"iterate feedback with twins\", \"keep iterating until\", \"loop twins on this\", \"converge on a variant\", \"/feedback-loops\".",
  "included_files": [],
  "skill_md_contents": "---\nname: feedback-loops\ndescription: >\n  Run an iterative feedback loop with August twins: put the same artifact\n  (copy, design, concept, variant set) in front of a panel of twins round\n  after round, applying the top suggestion from each round as a new variant,\n  until the responses converge on a preferred direction. Auto-invokes\n  /fix-question when a twin response is vague or hedged and re-asks that\n  twin with a sharper prompt. Meta-skill that orchestrates other August\n  skills so you don't have to manually kick each round. Triggers on: \"run\n  a feedback loop\", \"iterate feedback with twins\", \"keep iterating until\",\n  \"loop twins on this\", \"converge on a variant\", \"/feedback-loops\".\n---\n\n# Feedback Loops\n\nOrchestrate an N-round twin panel: same question, same twins, evolving\nartifact, until responses converge. Tools live under\nthe August MCP. This skill calls out to `/fix-question` for\nsharpening prompts mid-loop.\n\n> **Wall-clock note:** each round makes one MCP call per twin plus any\n> rephrase retries. Expect **several minutes per round**. Let the user know\n> up front so they don't ctrl-C mid-run.\n\n---\n\n## When to use\n\n- User wants to converge on the best variant of copy / a headline / a\n  concept, not just get a one-shot reaction.\n- User has an explicit stopping condition (\"until 4 of 5 twins pick the\n  same variant\", \"until reactions stabilize\").\n- User is willing to wait multiple rounds.\n\n## When NOT to use\n\n- Single-round gut check → `/august-feedback`.\n- Deep 1:1 discussion guide → `/august-interview`.\n- Comparing pre-existing variants without iteration → `/august-concept-compare`.\n\n---\n\n## STEP 1 — Nail the inputs\n\nConfirm you have:\n\n- **Artifact** — the concrete copy, design description, concept, or\n  variant set being tested. If vague, ask.\n- **Convergence criterion** — the explicit stopping rule. Common shapes:\n  - \"≥ K of N twins pick the same variant\"\n  - \"responses stabilize (no new objections in a round)\"\n  - \"top objection from round 1 is resolved\"\n- **Org + twin panel** — which org and which twins are on the panel. If\n  the user hasn't picked, use `list_organizations` + `list_twins` and\n  propose a 3–5 twin panel matching the target audience.\n\n**Hard cap: 4 rounds.** State this to the user at kickoff.\n\n---\n\n## STEP 2 — Round 1 (baseline)\n\nFor each twin in the panel:\n\n1. Call `start_interview` with its `twin_id` and retain the returned\n   `interview_id` for every round.\n2. Call `ask_question` with that `interview_id`, the question, and the\n   artifact in `design_context` when applicable.\n3. Capture the answer directly; call `get_interview_transcript` with the\n   interview ID as `conversation_id` only when the full thread is needed.\n\nReuse the same interview IDs in later rounds so each twin retains context.\n\nRecord for each twin: preferred variant (if applicable), top suggestion,\ntop objection, one-line summary.\n\n---\n\n## STEP 3 — Detect weak responses and repair\n\nAfter each round, scan every twin response for:\n\n- Vagueness (\"it's fine\", \"I guess it works\")\n- Hedging without content (\"kind of\", \"maybe\", \"depends\")\n- Off-topic drift (twin answered a different question)\n- Missing the concrete artifact (twin talked in generalities)\n\nFor any twin whose response fails this check:\n\n1. Invoke `/fix-question` internally with the original question. The skill\n   returns a sharper rewrite.\n2. Re-ask that twin using `ask_question` with the same `interview_id`.\n3. Replace the weak response with the repaired one.\n4. Cap re-asks at **1 per twin per round** — if it still comes back weak,\n   note \"twin response remained thin\" and move on.\n\n---\n\n## STEP 4 — Convergence check + next-round variant\n\nAfter every round, evaluate:\n\n- Does the panel meet the user's convergence criterion? → **STOP**, go to\n  STEP 5.\n- Are we at round 4? → **STOP** with \"hit hard cap without convergence\".\n- Otherwise → mint the next variant:\n  - Cluster the top suggestions from this round.\n  - Apply the highest-signal suggestion (most twins raised it, or the\n    strongest single suggestion) to produce a new variant of the artifact.\n  - Announce the change in the round trace (\"Round 3 variant: shortened\n    headline per twin B's suggestion\").\n- Loop back to STEP 2 with the new variant.\n\nDo NOT silently change more than one dimension per round — one variable\nper round keeps the signal clean.\n\n---\n\n## STEP 5 — Return the trace\n\nOutput structure:\n\n```\n# Feedback loop — {artifact name}\n\n*Panel:* {twin A}, {twin B}, … from {org}\n*Convergence criterion:* {criterion}\n*Rounds run:* {N} of 4\n*Outcome:* {converged | hit cap | stopped early}\n\n## Round-by-round\n\n### Round 1 — baseline\n- Variant: {description of artifact}\n- Responses:\n  - {twin A}: {one-line takeaway}\n  - {twin B}: …\n- Top suggestion applied to next round: {suggestion}\n\n### Round 2 — {what changed}\n…\n\n## Final winner\n{variant description}\n\n## Convergence commentary\n{2–4 sentences on how the panel moved across rounds — what shifted, what\nstayed stable, whether the panel actually agreed or just fatigued}\n\n## Suggested next move\n{one line — ship it, run another loop with a fresh panel, escalate to a\nfocus group, etc.}\n```\n\n---\n\n## Output rules\n\n- Always show the round-by-round trace, not just the final winner — the\n  journey is the signal.\n- Be honest about non-convergence. \"Panel did not converge; twin B held\n  out on X\" is more useful than a forced consensus.\n- Never fabricate a round. If a twin didn't respond or the MCP call failed,\n  say so in the trace.\n- If `/fix-question` had to be invoked, note it inline: \"(re-asked with\n  sharpened prompt)\".\n"
}