Personas: rework architect — mol-first, drop standing agent (tk-ae96t.2) — supersedes #132

[zook-bot]

Reworks the architect-persona deliverable per the operator's 06-17 review of #132 (28 inline comments) and the operator-confirmed direction. Supersedes #132 (bead tk-ae96t.2).

This is a fresh PR on polecat/tk-ae96t.2 rather than an in-place update of #132: the polecat work formula's submit gate requires the per-bead branch polecat/<bead-id>, so #132's polecat/tk-ae96t.1 head can't carry tk-ae96t.2's work. The diff is the complete reworked deliverable (it includes #132's commits + the rework), so #132 can be closed in favor of this.

Architectural changes (operator-confirmed)

1. Decouple architect-design / architect-review — moved from agent-local agents/architect/skills/ to the generic top-level skills/. They are now ordinary city-wide skills, invocable as mol steps (or when the identity is worn), no longer bound to a standing agent. Resolves the SKILL.md disconnect ("method-skills ride with the standing agent, so a transient wearer brings the lens, not those skills" vs. a mol step that includes architect-review).
2. Drop the standing architect agent — removed agents/architect/ (agent.toml, prompt.template.md, PROVENANCE.md) and the pack.toml [[named_session]]. No formula, order, keybinding, or city.toml import referenced it, so the removal is load-safe.
3. Add the proof-point mol — formulas/mol-architect-review.toml runs architect-review as a mol step: the first proof the method works with no standing agent in the picture.

Resolution of #132 review comments

agents/architect/* — resolved structurally by removing the standing agent

• agent.toml L6, L10, L12, L21 (comment bloat; "skeptical this should be in Gas City at all / should be rig-scoped, instantiated only for gc-toolkit; we don't know how it's used") → file removed; there is no standing instance to describe.
• prompt.template.md L7, L14, L21, L31, L37, L47, L60 (unnecessary / duplicated identity & method detail that belongs in the skill, not the prompt) → file removed.
• PROVENANCE.md L51 ("why put something in a document that is invalidated upon review and then needs to change?") → file removed (the agent it documented is gone).

docs/personas.md

• L9 "are now settled" → removed; no status-meta (docs/* is authoritative by definition).
• L23 "eg: architect-review" → the Methods bullet now uses the concrete architect-review example.
• L49 "(red-green)" → removed; the section states the goal (a method available wherever the work is, without bloating every session), not specifics.
• L59 multi-framework caveat → removed (we use the generic skills directory).
• L64 persona↔bead / triage detail → removed from the Orchestration layer.
• L83 "this sketch is now built…" → removed; the concrete instantiation now lives in a factual "In Gas City: the architect" section, not a status narrative.
• L88 "Mechanics (settled)" → section replaced by a "Persona structure" table; no "settled" framing, no implication that part of the doc carries different weight.
• L93 → rewritten as a persona-structure definition (table: path → what it is → what it does), far less prose.
• L117 closing retrospective ("the build added one lesson worth keeping…") → removed.

skills/architect/SKILL.md

• L3 description → dropped the "wear this to reason in its two modes" agent-choice framing and the "method-skills ride with the standing agent" claim; methods are now independent skills.
• L8 → replaced the verbose, process-centric intro with a direct identity opener: "You are now the architect. What follows defines your identity as the architect."
• L17 tone → reframed to shepherd/lead: guide, ensure consistency, help others see how their work fits the bigger architecture, lead where it's headed — not "ask more questions."
• L22 "This is great" → "What I optimize for" kept verbatim.
• L41, L71 ("same architectural issue around personas") → methods / engagement sections reworked: methods are ordinary skills (mol step or worn identity); a standing agent is an optional future, not the carrier of the methods.

pack.toml

• L100 ("tempted to leave out the agent instance completely and focus on a mol that leverages the architect as the first proof point") → the architect [[named_session]] is removed; the architect is proven via mol-architect-review.

Reconciliation (not directly commented, but made consistent)

• docs/roadmap.md — the "revived as a persona" entries no longer claim a standing agent ships; they describe the skills + the mol-step proof, with a standing agent deferred until drift-patrol / structural gating is actually needed.
• README.md — persona index line: "process-skills" → "method-skills".

Out of scope (the host handles)

• Closing Land docs/personas.md — persona/agent model (persona = skill) (tk-oe8o0) #123 / Personas: settle the mechanics (skill load paths, subagent loading, scoping) (tk-ohrlc) #130 (superseded earlier) and Personas: first architect persona (BMAD-grounded) on Path A — supersedes #123 & #130 (tk-ae96t.1) #132 (superseded by this).

tk-ae96t.2

[zook-bot]

Blocking finding:

• specs/tk-ae96t.1/README.md:37 still documents the old tk-ae96t.1 implementation as if it is what this PR lands: method skills under agents/architect/skills/, an agents/architect/{agent.toml,prompt.template.md,PROVENANCE.md} standing-agent form, and a pack.toml named-session registration. In this PR's final tree those paths are absent; the actual delivered shape is top-level skills/architect* plus mol-architect-review, with no standing architect agent. Because this README is newly added by the PR and docs/personas.md points readers at specs/tk-ae96t.1/ for the architect grounding, merging as-is would leave a repository record that contradicts the current persona contract and revives exactly the old shape the PR says it dropped. Please update this record to mark tk-ae96t.1 as superseded by tk-ae96t.2 and describe the final mol-first/top-level-skill shape, or clearly isolate the stale Path A details as historical pre-rework notes.

Validation: checked out PR #166 detached, parsed formulas/mol-architect-review.toml with tomllib, ran git diff --check origin/main...HEAD, and verified the claimed agents/architect paths are absent from the final tree.

Note: GitHub would not accept this as a formal REQUEST_CHANGES review because this bot account owns the PR branch, so I am posting it as a blocking review comment and routing the fix through the bead workflow.

[zook-bot]

Re-review after fix 60f8f9f: no blocking findings.

The previous blocker in specs/tk-ae96t.1/README.md is addressed. The file now marks tk-ae96t.1 as superseded by tk-ae96t.2, lists the final mol-first/top-level skill shape, and clearly fences the old agents/architect/... and pack.toml entries as Path A historical/pre-rework notes.

Validation:

• checked PR head 60f8f9f against origin/main
• confirmed agents/architect/ is absent from the final tree
• parsed formulas/mol-architect-review.toml and pack.toml with python3/tomllib
• ran git diff --check origin/main...HEAD

[zook-bot]

Blocking finding:

• specs/tk-ae96t.1/research/architect-prior-art.md:12 still describes the architect persona as including skills/architect, skills/architect-design, skills/architect-review, and the agents/architect standing form, then says that shape "is built." The latest reconciliation fixed the top-level specs/tk-ae96t.1/README.md banner and the docs/personas.md pointer, but this newly added research index is the linked prior-art/distillation record under that same directory and has no superseded/Path A historical framing of its own. A reader following the record still lands on a current-tense statement that the absent standing-agent form exists, contradicting the final PR shape (skills/architect* + formulas/mol-architect-review.toml, no agents/architect/). Please either add the same tk-ae96t.2 supersession/historical framing to this index or rewrite the sentence so it names the final mol-first/top-level-skill shape.

Validation:

• checked PR head e904bc5 against origin/main
• confirmed the latest reconciliation commit updated docs/personas.md, specs/tk-oe8o0/README.md, and specs/tk-ohrlc/README.md
• parsed formulas/mol-architect-review.toml and pack.toml with python3/tomllib
• ran git diff --check origin/main...HEAD
• gh pr checks 166 reports no checks on the branch

Note: GitHub would not accept this as a formal REQUEST_CHANGES review because this bot account owns the PR branch, so I am posting it as a blocking review comment and routing the fix through the bead workflow.

[zook-bot]

Re-review after fix dba41c2: no blocking findings.

The previous blocker in specs/tk-ae96t.1/research/architect-prior-art.md is addressed. The file now marks the prior-art index as superseded by tk-ae96t.2, frames Path A's standing architect agent / agent-local method-skills as historical, and names the final shipped shape as top-level city-wide skills/architect, skills/architect-design, skills/architect-review plus formulas/mol-architect-review.toml, with no agents/architect/ standing agent.

Additional checks:

• sibling spec records and docs/personas.md point readers at the mol-first/top-level skill shape rather than the old Path A shape;
• agents/architect is absent and pack.toml does not add an architect named session;
• the new formula file is valid TOML with the expected three steps;
• the PR head is fast-forwardable from current origin/main.