[ffroliva]

Summary

The HAR/DOM harness landed in #703, but nothing routed to it — so an agent still had to stumble on it. That is the failure that cost a day: a working feature was declared impossible because nobody measured it, and the harness that would have settled it in thirty minutes sat undiscovered.

This makes recon a phase, not a tool.

• skills/spike/SKILL.md — canonical, agent-agnostic (Codex / Cursor / Aider read the same file), plus the /gflow:spike wrapper
• AGENTS.md Skill Routing — added as the first row, because it gates the others
• Standard Workflow Sequence — new Phase 0: Recon, before Triage
• issue-assessment — a "Before you classify an absence" section. A verdict of INVALID / "not supported on this host" is a claim about the live product, and that skill is read-only, so it cannot produce one from the code alone. Closing a reporter's issue on an unmeasured negative tells a user their working feature is impossible.

The rule

A selector that does not match is evidence about the SELECTOR.
It is never evidence about the FEATURE.
Never put a guard in front of a probe.

Both halves come from the same failure. "character create is a labs-only surface — renders no prompt textbox for it, ever" came from one 20 s selector timeout and reached four places: a code comment, the CHANGELOG, the v0.70.0 release ledger, and a test class name. #701 then added a guard that aborted before probing the DOM — which made the claim unfalsifiable, because no run could ever look.

New finding recorded, not yet acted on

docs/superpowers/spikes/2026-09-06-migrated-character-surface-and-recaptcha.md records the full recon behind #703, plus something new:

flow.google.com CAN mint a reCAPTCHA Enterprise token.

route	recaptcha scripts	site key	grecaptcha.enterprise	mint
flow.google.com/ (root grid)	0	—	false	❌ RecaptchaError
flow.google.com/project/<id>	2	6LdsFiUsAAAA…	true	✅ OK, 2404-char token

The mint guard refuses the whole host when the pooled bootstrap page is simply on the wrong route. That is the mechanism behind #692 — whose reporter's bundle showed route: "/", and diagnostics.py maps both hosts to host_category: flow_app, so the route is the only tell.

Deliberately not claimed: that fixing the mint unblocks gflow image. That path is UI-driven (mint → drive the composer), and the migrated project composer inventories as video-only — movie, apps_spark_2, crop_16_9, [role=tab] = 0, no image-mode control. Image generation exists on that host only inside the character editor. So the mint is necessary, not sufficient, and "one anchor sweep away" is recorded as unproven.

Test plan

• All six doc/hygiene gates green (check_doc_links, check_repo_hygiene over 1008 files, check_council_memory, generate_website_docs --check, check_website_docs_pii, ruff)
• 43 documentation/skill/agents gate tests pass
• scripts/dev/spike_migrated_recaptcha_mint.py is committed and re-runnable; its evidence is gitignored
• Docs-only plus one spike script — no Flow surface touched, so no e2e applies (AGENTS.md's explicit out-of-scope case)