tiktok-architect: decode the ironic 'never ask Claude' format (verified live 2026-07-19) — flip before scoring, harden safety-gate framing

434fa7c7c07b883e23e19827cd45eea99c1f84b8 · 2026-07-19 07:27:36 -0700 · Steve Abrams
Files touched

M SKILL.md
M references/sources.md
Diff

commit 434fa7c7c07b883e23e19827cd45eea99c1f84b8
Author: Steve Abrams <steve@designerwallcoverings.com>
Date:   Sun Jul 19 07:27:36 2026 -0700

    tiktok-architect: decode the ironic 'never ask Claude' format (verified live 2026-07-19) — flip before scoring, harden safety-gate framing
---
 SKILL.md              | 15 ++++++++++-----
 references/sources.md | 25 +++++++++++++++++++++----
 2 files changed, 31 insertions(+), 9 deletions(-)

diff --git a/SKILL.md b/SKILL.md
index 83436ef..2fe93f8 100644
--- a/SKILL.md
+++ b/SKILL.md
@@ -54,11 +54,16 @@ The feed is **mixed** — motorcycles, hobbies, privacy, AND Claude tips. Scout
 **only** the videos where `claude_relevant` is true. Do not treat every post as
 a tip. (See the mixed-feed note in `references/sources.md`.)
 
-### 3. Extract the candidate tip
-For each relevant video capture: **claim** (the assertion, restated as an
-imperative rule), **kind** (anti-pattern / best-practice / workflow / tool-repo /
-prompt-pack / setting), and the **specifics** (the *why* + the *instead-do-this*)
-from caption/on-screen/transcript. If the tip names a repo, resolve + enrich it:
+### 3. Extract the candidate tip — DECODE THE IRONY FIRST
+@tiktok_architect's *"Never ask Claude to do X"* is **reverse psychology**: it
+means *"here's an X you can build."* Flip every "never/don't" to "here's how to"
+and evaluate the REAL payload — never the literal surface. (Verified live
+2026-07-19; the decoded payloads trend TOS-violating / harmful, so most land at
+SKIP. See `references/sources.md`.) For each relevant video capture: **decoded
+claim** (the real thing being taught, as an imperative), **kind** (anti-pattern /
+best-practice / workflow / tool-repo / prompt-pack / setting), and the
+**specifics** (the *why* + the *how*) from caption/on-screen/transcript. If the
+tip names a repo, resolve + enrich it:
 ```bash
 node scripts/gh-repo-stats.js owner/repo
 ```
diff --git a/references/sources.md b/references/sources.md
index 1a2e491..5d643f4 100644
--- a/references/sources.md
+++ b/references/sources.md
@@ -20,11 +20,28 @@ Claude/AI-coding videos.** `tt-list.js` sets `claude_relevant` per video by
 hashtag/keyword — trust that flag and skip the rest. Never treat every post as a
 tip.
 
-## ⚠️ Tips are CLAIMS, not facts — verify before codifying
+## ⚠️ The "Never ask Claude for X" format is IRONIC — decode it FIRST
 
-This is the core difference from a tool-scout. A behavioral tip ("never ask
-Claude Code to X") is engagement-optimized and frequently **outdated, model/
-version-stale, or just wrong**. The rubric's first gate is **correctness** — a
+Verified live 2026-07-19 by transcribing 5 videos: this creator's whole format is
+**reverse psychology**. *"Never ask Claude to do X"* literally means *"here's an X
+you can build with Claude."* The sarcasm ("we definitely wouldn't want to break
+terms of service, right?") IS the gag. **Before evaluating any tip, decode the
+irony** — flip every "never/don't" to "here's how to", and judge the REAL payload.
+
+And the real payloads trend **harmful**, not helpful: the sampled run decoded to a
+FB-Marketplace→eBay arbitrage scraper (TOS), a game aimbot/ESP wall-hack site +
+AdSense/SEO spam (cheat + ad fraud), a WebRTC chat client that logs peer IPs to
+"bait creeps" (entrapment/privacy), and an uncensored-LLM chat bot posing as an
+18-19yo for adult affiliate offers (illegal). So this is **NOT a Claude-Code
+best-practices feed** — it's edgy engagement bait. Expect the safety dimension to
+route most/all decoded tips to SKIP, and flag the harmful ones loudly. NEVER build
+one of these payloads; the scout only surfaces + classifies them.
+
+## ⚠️ Tips are CLAIMS, not facts — verify (after decoding the irony)
+
+A decoded tip is still just a CLAIM. Even the rare genuinely-useful one is
+engagement-optimized and frequently **outdated, model/version-stale, or just
+wrong**. The rubric's first gate is **correctness** — a
 tip that isn't true for *current* Claude Code is SKIP no matter how catchy.
 Verify against the `claude-code-guide` subagent, the `claude-docs` / `claude-api`
 skills, and official docs (docs.claude.com / docs.anthropic.com) BEFORE it is
← eb4b7f3 tiktok-architect scout skill — initial build · back to Tiktok Architect · runs: capture 2026-07-19 live scout report + decoded transcr 85a79e2 →