plans: ck plugin phase-one PRD, revision 3, with Phase 0 spike results

[d6veteran]

Summary

Turns plans/2026-09-05-agent-workflows-research-and-proposal.md (committed here alongside) into a phase-one PRD for a single code-katz plugin named ck, revised three times (first draft; a cross-model panel against an Opus-written PRD; a rewrite after Will's sixteen-comment review), and now carries the results of the Phase 0 spikes.

Current state: ck is an independent replacement for the old team tool. Phase one is the full product-definition pipeline plus a design step, /ck:panel, /ck:next, and 21 personas as subagents and switch commands on three model tiers. /ck:feature is confirmed for phase two. The Workbench is phase three. Six of eight Phase 0 spikes are answered; two need a local machine.

Commit 6: Phase 0 spikes

A throwaway ck plugin (plans/phase-0/ck-spike/) was loaded into a nested Claude Code with --plugin-dir and driven non-interactively. Record: plans/2026-09-09-ck-phase-0-spikes.md.

• Yes: a persona agent inside a workflow runs on its frontmatter model and a per-call model overrides it (S2); the SubagentStart hook fires for direct delegation and inside workflows with agent_type, agent_id, session_id, cwd, and writes under CLAUDE_PLUGIN_DATA (S4); a neutral agent inside a nested workflow ran a real web search (S7); /ck:next registers and ${CLAUDE_PLUGIN_ROOT} expands (S1, this machine); the extra personas key in meta loads; a plugin workflow is addressed by name: "ck:<workflow>".
• One design change: a nested workflow is called by name, workflow('ck:panel', args); calling by script path is refused when the plugin is outside the working directory. draft.js, the /ck:prd skill, and §6.6 updated.
• From the docs (S5): review pages work in the CLI (2.1.183 or later) and the desktop app on any paid plan; comments need 2.1.221 or later and a Team or Enterprise organization. Clare's plan is the check; the file-edit fallback stays at every gate.
• Still local: the interactive consent prompt for a nested workflow (S3), a marketplace install on a second machine (S1), Clare's CPU count (S8). The drill log has the commands.

Commit 5: /ck:feature confirmed for phase two

Will answered the one open scope question on 2026-09-09. PRD §1, §2.5, §3.2, §3.4, §10.2, and §10.4 record it.

Commit 4: revision 3 (Will's review, 2026-09-08)

Every comment and the decision it drove is recorded in PRD §3.4. The changes:

• ck is independent. No coexistence, no retirement debate, no vendored profiles. ck owns profiles/ and tiers.conf after a one-time import; uninstalling the old tool is a prerequisite, with a SessionStart hook that warns while it remains.
• Clare is the primary user, proficient with the current tool and running many sessions. Two journeys: J1, an idea to a full definition; J2, one feature to design mockups, the same way every time. Efficiency (fewer sessions and hand-offs) is a measured goal.
• Paths: <project-repo> is defined once; every document is written into the product's own repository under docs/ (or ROADMAP.md, brand/) and committed.
• Models: Fable 5.1 for the six judgment seats (River, Akira, Morgan, Sage, Jordan, Reiner), Opus 5 for the eleven craft seats, Sonnet 5 for the four execution seats, Haiku for validators. The panel default is River on Fable, Toni on Opus, Kai moved down to Sonnet.
• Phase one is the pipeline: /ck:opportunity, /ck:market-research, /ck:brief (with a market pass), /ck:prd, /ck:team (roles and responsibilities), /ck:roadmap, /ck:architecture, /ck:brand-guide (proposals, finalists, guide, modelled on the NIGHTGRID process and the d20Mob guide), /ck:design. Ten document contracts. One drafting engine (draft.js) serves the PRD and the architecture document.
• Reviews go through Claude's built-in review and comment system; every review page carries the five how-to-comment steps in its banner.
• Phase two: /ck:feature, /ck:bugfix, /ck:gtm, Routines, /ck:map, /ck:report. Phase three: the Workbench, a rewrite replacing the conductor dashboard, local, never a hosted store.

Earlier commits

• Commit 1, the PRD. Proposal §8 resolved first: a sign-off is a gate in a skill; each span between gates that needs fan-out is one plugin workflow; a single-agent span runs inline. Corrections to the proposal, each graded. Plugin vs web app: plugin.
• Commit 2, the cross-model revision. Nine changes adopted from the Opus PRD and eleven rejected with verified reasons. Record: plans/2026-09-05-ck-prd-panel-memo.md, with §8 for the items Will's review superseded.
• Commit 3. A one-line count fix in the PRD introduction.

Appendices

Four complete workflow scripts (panel, draft, brief, team; syntax-checked as the runtime runs them, body wrapped as an async function), two skills (prd, next), the brief and PRD contracts, the generated River agent excerpt, a cost model per command at 2026-09-05 prices, and sources. The other five workflows are specified to the stage level in §6.

Also in this PR

• DEVLOG.md: four entries, one per revision and one for Phase 0.
• plans/2026-09-05-agent-workflows-research-and-proposal.md and plans/opus/: the inputs, verbatim.
• plans/phase-0/: the spike plugin and trimmed run results.

Not in this PR

No product code. The code-katz/ck repository and the marketplace entry are the build, which starts from this PRD after the two local spikes are recorded.

🤖 Generated with Claude Code

https://claude.ai/code/session_014aqcjmXas9EFDx3pstfpyE