Review of the spec at https://repo-reorg.askelventures.com/ Date: 2026-05-13
Phases 1 and 3 are executable as written. Phases 2 and 4 are not — they'd stall or go wrong without additional inputs. There are also a handful of systematic gaps worth fixing before anyone fires execute.
1. Every Drive document link is a placeholder.
The "Drive Linking — The Rule" section declares that every mention must be a hyperlink, then every entry in the inventory table says [open]. An executor writing Phase 2 skills would be unable to comply with the rule it's enforcing. Before Phase 2 starts, someone needs to run gws drive list (or open Drive), resolve the six canonical URLs, and paste them into the spec. Without this, Phase 2 produces skills that are immediately non-compliant.
2. Case Scouting board URL is nowhere in the spec.
The Trello TEMPLATE card URL (https://trello.com/c/ZWY6fNrB) survives from CLAUDE.md. The Case Scouting board URL doesn't appear anywhere — not in the spec, not in CLAUDE.md. Phase 1 explicitly says "add Case Scouting board and template card links inline" to three cron skills. One of those two links is unresolvable without someone going to Trello and grabbing it.
3. cron-yrityskaupat-newsletter and cron-walk-of-shame fall through the cracks. The spec's own external references gap analysis flags both as needing Case Scouting board links. Neither appears in Phase 1's "files to modify" list. They're identified as broken, then not fixed.
4. Estonia is added to the thesis but has no sourcing cron.
Geography expands to Finland + Sweden + Estonia. Sweden has cron-swedish-sweep. Estonia has nothing. The spec notes the geography change, updates cron-swedish-sweep to acknowledge Sweden (which was already its mandate), but never addresses whether Estonia gets folded into the Swedish cron, gets a new cron, or is deliberately left as manual-only. This is a logic hole in the hypothesis coverage.
5. git mv will fail without mkdir first.
knowledge/portfolio/ doesn't exist. git mv knowledge/melers knowledge/portfolio/melers/brand will error because the parent directory isn't there. The fix is a mkdir -p knowledge/portfolio line before the mv. Small thing that kills Phase 3 immediately.
6. "One full sprint" is undefined. The deprecation timeline says delete sourcing-tradecraft "after one full sprint with no incidents." Sprint length is never defined anywhere. Is that a week? Two? A month? An executor has no anchor.
7. Phase 4 confidentiality mechanism is a hand-wave. "Mark confidential" appears as a bullet, with no specification — no password, no access control, no header convention. The askel-pages skill presumably publishes publicly. An executor would have to invent the mechanism.
8. Phase 4 publishing format is unspecified. "Generate playbook.askelventures.com from skill markdown files" — how? Which skills? In what order? With what navigation? The askel-pages skill gets named but its invocation pattern isn't described. This entire phase needs a design doc or at least a paragraph of prose.
9. CLAUDE.md update scope is underspecified in Phase 2.
Phase 2 says update the knowledge base index in CLAUDE.md. But CLAUDE.md also has a #prospecting protocol section that hardcodes hypothesis descriptions and cron names, and a ## Your jobs section that references sourcing directly. These probably need touching too, but the spec doesn't mention them.
10. GOG_ACCOUNT env var is mentioned once and then never referenced. The execution context says "set GOG_ACCOUNT env var before any gws call." No phase instruction actually invokes gws. If an executor needs to resolve Drive URLs (which they do, per gap #1), they'd need gws — and they'd find the env var underdocumented.
- The
playbookrouter skill says "reads context, surfaces the relevant sub-skill" but trigger conditions are left for the executor to invent. Defensible for a router, but a one-liner per sub-skill on what triggers it would help. - The "After" directory tree shows
sourcing-tradecraft/still present (correct — it's in deprecation limbo), but it'll look like a mistake to anyone who only reads the tree and not the surrounding text.
Phase 1 is tight. Files identified, changes described precisely, the thesis merge strategy is clear, the conflict resolution rule (v2 wins) is explicit. Phase 3 is close — just needs the mkdir fix. The deprecation pattern, skill format spec, and canonical source selection are all unambiguous.
The spec is ~80% of the way there. The Drive URLs and Case Scouting board URL are the only things that would actually stall work. Everything else is friction, not a full stop.