Signal Skills: ideation
Turn findings already mined by podcast/review-scout into prioritized "build/test next" proposals, each citing the specific finding(s) that motivated it. Use when the user asks "what should we build next", "what's worth prioritizing", or wants a synthesis across research + reviews instead of another raw findings dump.
Findings are not decisions. podcast and review-scout each produce ranked
findings; this skill is the one step further — turning "23 reviews complain
about slow checkout" into "ship a one-tap checkout retry, here's why, here's
the confidence." Every proposal must cite a finding. An idea with no
citation is a guess wearing a findings hat — don't generate those.
Step 0 — Check the ledger before proposing
python3 ~/.claude/skills/ideation/ledger.py list --status proposed
python3 ~/.claude/skills/ideation/ledger.py list --status shipped
Don't re-propose something already proposed (still pending a decision),
shipped (done), or rejected (already said no, would need new evidence to
revisit). Only parked ideas are fair game to resurface if new findings
strengthen the case.
Step 1 — Pull source findings
Read the other skills' state directly, don't re-derive from scratch:
python3 ~/.claude/skills/podcast/ledger.py known-issues 2>/dev/null
python3 ~/.claude/skills/review-scout/ledger.py known-issues 2>/dev/null
python3 ~/.claude/skills/review-scout/ledger.py dedup --threshold 0.5 2>/dev/null
If the user already has a specific rendered report (podcast/review-scout output) in this conversation, use that directly instead of re-querying the ledger — it's the same data, already synthesized.
Step 2 — Generate proposals, one per candidate
For each build/test candidate, work out:
- The citation — which specific finding(s) motivate this, verbatim claim text or a short paraphrase plus source (app/show, batch/episode).
- The shape — is this a bug fix, a feature, an experiment, or a "needs more data" flag? Don't force everything into "build this."
- Effort vs. confidence — cheap+high-confidence beats expensive+ speculative; say which bucket this falls in, don't just list ideas flat.
- What would make this wrong — the one thing that, if true, kills the idea. Forces you to actually think about it instead of pattern-matching "reviews complained → build the fix."
Record each as you go:
python3 ~/.claude/skills/ideation/ledger.py propose --title "<short title>" \
--citations-json '["23 reviews cite slow checkout, v2.4-2.6 (review-scout)", "..."]' \
--note "<effort/confidence/kill-condition, one line each>"
Step 3 — Rank, don't just list
Sort by confidence × leverage, not by recency or by which source skill produced it. A single well-corroborated review-scout regression outranks three speculative podcast takeaways with no repeat citation.
Step 4 — Report (shared renderer)
Same render_report.py/template.html as the other three skills
(a real copy here, kept in sync with the canonical one in signal-report
— packaged plugins ship real files, not symlinks, so this installs standalone). Tiers should reflect decision-readiness, not just
severity — e.g. "Ship now (cheap, confirmed)", "Worth a spike (confidence
gap)", "Needs more data before deciding", "Parked (rejected once, revisit
if new evidence)".
python3 ~/.claude/skills/ideation/render_report.py payload.json out.html
Publish with Artifact on Claude Code; elsewhere use --open or report
the path.
For a paste-into-a-ticket/prompt form of the same proposals, add a .md
target in the same invocation:
python3 ~/.claude/skills/ideation/render_report.py payload.json out.html out.md
Step 5 — After the user decides
Update status so the next run doesn't re-propose it:
python3 ~/.claude/skills/ideation/ledger.py set-status --title "<title>" --status shipped
python3 ~/.claude/skills/ideation/ledger.py set-status --title "<title>" --status rejected
python3 ~/.claude/skills/ideation/ledger.py set-status --title "<title>" --status parked
Hygiene
- State lives in
~/.claude/skills/ideation/config.jsonandstate/ideas.json— a real copy of the ledger pattern, deliberately NOT symlinked to podcast's or review-scout's ledger.py (samePath(__file__).resolve()-follows-symlinks state-isolation bug that ruled out sharing ledger.py anywhere else in this skill family). - This skill reads podcast/review-scout state but never writes to it.
- Downloads
- Security audit
- Pass
- Last updated
- 1w ago
- Current version
- v1.0.0
- License
- MIT-0
