Synced from
references/virality-bridge.mdinscript-poolskill on 2026-05-18. Edit upstream in the skill; this file is overwritten on next sync.
References Bridge — to afriend-content-extraction canon
This skill does NOT duplicate the virality canon. It reads from afriend-content-extraction/. This file maps which sub-files apply to script-writing (pre-recording, the script-pool’s job) vs recording-extraction (post-recording, the other skill’s job).
Read by Mode C (draft-script)
| Path | Why used here |
|---|---|
~/.claude/skills/afriend-content-extraction/references/shortform-virality.md | Tier-1/2/3 hook patterns (which patterns to compose draft against), arc shapes (cold-open, vulnerability-then-recovery, etc.), Tier-1 self-supplying scroll-stopper rules |
~/.claude/skills/afriend-content-extraction/references/attention-preservation.md | Retention windows (0-1.5s 80%, 12-18s second-hook critical, 30-45s resolution start), anti-brain-rot definition (lines 176-192) |
~/.claude/skills/afriend-content-extraction/references/essence-extraction-framework.md | PARTIAL — kernel test (“if this clip had to fit one Daniel-sentence, which?”), compression math caps. NOT used: Schärfung trim phase (post-record only), Trim 2a/2b weakness elimination (post-record only) |
~/.claude/skills/afriend-content-extraction/priming/tool-naming-checklist.md | Pillar-7 verification — 4-question drill validates that script body names tools + combination + low-stakes verb |
Read by voice-safety gates
See references/voice-safety-gates.md for full gate-to-source mapping. Brief:
- Gate 6 (tool-naming-when-pillar-7) → reads
priming/tool-naming-checklist.mdwhitelist - Gate 7 (anti-brain-rot) → reads
references/attention-preservation.mdlines 176-192
NOT read
The virality stages are recording-extraction-only and don’t apply to script writing:
| Path | Why NOT used |
|---|---|
~/.claude/skills/afriend-content-extraction/virality/SKILL.md | Orchestrator for S1/S2/S3 stages over EXISTING transcripts |
~/.claude/skills/afriend-content-extraction/virality/stages/01-source-analysis.md | Inventories beats from transcript silence detection — no transcript exists pre-recording |
~/.claude/skills/afriend-content-extraction/virality/stages/02-candidates.md | Generates 5-7 candidate clips from existing inventory — N/A pre-recording |
~/.claude/skills/afriend-content-extraction/virality/stages/03-refinement.md | Composes cut-plan.json from selection — N/A, the script IS the plan |
If a future feature wants to “post-validate” a script against the virality canon AS IF it were a recorded transcript, the bridge here would expand. Today: not needed.
Cross-skill discipline
- Script-pool only reads from afriend-content-extraction. Never writes there. Exception: the one-time additive edit to
priming/SKILL.md(script-driven branch) is a deliberate cross-skill modification documented in M113 S02.8. - If a virality canon line changes upstream, the change automatically flows here on next read — no copy-paste, no sync drift.
- If script-pool needs new canon (e.g., educational-format-specific virality patterns), propose adding to
afriend-content-extraction/references/rather than duplicating in script-pool. Single source of truth.