consult-resume

Engagement Re-entry

Re-enter a cogni-consult engagement: discover what exists, show progress against the action-fields WBS (fields × deliverables × status), and route to the most valuable next action. This skill is a read-only orienter — it never edits engagement state; the only files it writes are the derived front-door artifacts (README.md, assumptions.md) regenerated from that state, and every other write belongs to the skill it routes to.

Workflow

1. Discover Engagements

bash $CLAUDE_PLUGIN_ROOT/scripts/discover-projects.sh --json

When discovery returns zero engagements, there is nothing to resume here. If the user was working a phase-based (Double Diamond) engagement, that engagement model is no longer part of the ecosystem — those engagements live in git history. Otherwise recommend scaffolding an engagement and dispatch Skill("cogni-consult:consult-setup"), then stop — setup owns scaffolding and the knowledge-base binding.

2. Select the Engagement

• One engagement → select it silently.
• Multiple, and the user named one → fuzzy-match the named name or slug against discovery and select it directly; confirm only when the match is ambiguous.
• Multiple, and the user named none → you MUST output the full engagement list (name, slug, scope_state, scope config updated) and STOP for the user's explicit choice. Never silently select or infer an engagement in this case. The active git branch, working-tree / uncommitted changes, and recent commit history are not authorized selection signals — they must never substitute for the user's choice or override the list-and-stop obligation. Only a name or slug explicitly named in the user's message may skip the list.

Conduct the conversation in the resolved interaction language (workspace default, overridden by the user's message language) — independent of the engagement's language field, which is the deliverable axis. See $CLAUDE_PLUGIN_ROOT/references/interaction-language.md.

3. Read the Engagement Status

bash $CLAUDE_PLUGIN_ROOT/scripts/engagement-status.sh <engagement-path>

<engagement-path> is the path field from discovery. The script derives the rollups at read time: scope_state, an engagement-level personas_gate (satisfied once a scope-seeded persona or a .gate-waiver marker exists, else pending), and per field its state plus each deliverable's state, dt_stage, producing_route, and persona_review. Surface any warnings[] verbatim (an unreadable field manifest is a problem for the consultant to see, not to paper over).

4. Present the Dashboard

Lead with the key question, then one table row per action field:

Engagement: <name> (<slug>) — scope config updated <date>
Key question: <key_question>

| Action Field | Status | Deliverables | Next Deliverable |
|--------------|--------|--------------|------------------|
| market-evidence | complete | 2/2 complete | — |
| portfolio-fit | in-progress | 1/3 complete | competitor-map (ideate · pyramid-principle) |
| go-to-market | pending | 0/2 started | channel-strategy (empathize · —) |

Deliverables counts complete over total; Next Deliverable names the first non-complete deliverable with its dt_stage and stored chosen_framework in parentheses (<stage> · <framework>), or the first whose persona_review is still open when everything else is done. The framework is surfaced read-only — a registry slug verbatim, or for a combo:<slugA>+<slugB> pairing the two slugs joined as <slugA> + <slugB> (the stored combo: prefix dropped for display), or — when none is stored (legacy deliverables); it is never inferred here. Keep it to this one table — the deep WBS view (planning deliverable sets, splitting fields) belongs to consult-action-fields, not here.

Offer the visual dashboard. After the text table, offer the consultant a themed, browsable HTML view of the same status via /cogni-consult:consult-dashboard (action-field WBS, deliverable states, design-thinking stages, persona-review coverage). When the engagement already has output/design-variables.json from a prior dashboard run, you can regenerate and open it without a theme prompt by delegating to the consult-dashboard-refresher agent with engagement_dir: <engagement-dir> and plugin_root: $CLAUDE_PLUGIN_ROOT. This stays read-only — the agent runs the read-only generator; it never edits engagement state.

Point at the project plan (only when scheduling data exists). When the engagement carries a schedule, surface it here as a one-line read-only pointer. Detect it cheaply: a <engagement-dir>/project-plan.md already exists, or python3 $CLAUDE_PLUGIN_ROOT/scripts/deliverable-graph.py "<engagement-dir>" schedule returns a non-empty data.schedule[] with at least one scheduled (not unscheduled) entry. When it does, name the next critical-path deliverable — the first not-yet-complete deliverable whose key is in data.critical_path[] — and offer /cogni-consult:consult-project-plan to (re)render project-plan.md (phase timeline + critical path + optional gantt). When no deliverable carries scheduling fields (the schedule read is empty or every entry is unscheduled), say nothing — this pointer degrades silently and the dashboard above is unchanged. It is informational only: do not add it as a branch in the Step-5 next-action ladder — the project plan is a derived read model, not a competing recommendation.

Milestone README refresh. On every re-entry, also run python3 $CLAUDE_PLUGIN_ROOT/scripts/generate-engagement-readme.py "<engagement-dir>" automatically (no prompt) to refresh the engagement-root README front door — unconditional (unlike the theme-gated dashboard offer above, no output/design-variables.json needed) and non-fatal: on failure, warn and continue. An engagement that predates the README front door gains one here on its next re-entry — no migration step needed; a hand-authored root README.md is never overwritten (the generator refuses when its marker footer is absent — that refusal is the same non-fatal warning case). The README is a derived front-door artifact regenerated from engagement state, not engagement state itself, so the read-only contract over consult-project.json, field.json, personas, and logs holds.

Assumption register refresh. On every re-entry, also run python3 $CLAUDE_PLUGIN_ROOT/scripts/register-generator.py "<engagement-dir>" automatically (no prompt) to regenerate the engagement-root assumptions.md register — the human-browsable table of every registered assumption's value, provenance, status, and used_by[] backlinks — from assumptions.json. Unconditional and non-fatal: the generator's load_assumptions is fail-soft (a missing, empty, or malformed registry degrades to an empty register, never an error), and on any failure, warn and continue. An engagement that carries a populated assumptions.json but never had its readable register generated gains one here on its next re-entry — no migration step needed; a hand-authored assumptions.md is never overwritten (the generator refuses when its generated marker footer is absent — that refusal is the same non-fatal warning case). Like the README, assumptions.md is a derived read model regenerated from the registry, not engagement state itself, so the read-only contract above holds.

Point at the register (only when it is non-empty). When the generator returns success: true with data.assumptions_count > 0, surface a one-line read-only pointer to the regenerated assumptions.md — mirroring the project-plan pointer above so the refresh is visible rather than silent (e.g. "Assumption register: assumptions.md (N registered) — value, provenance, and used_by[] backlinks"). When the count is 0 (no registered assumptions) or the generator warned, say nothing — the pointer degrades silently. It is informational only: like the project-plan pointer, do not add it as a branch in the Step-5 next-action ladder.

Strategy Advisor voice — this plugin ships the Strategy Advisor output style (answer-first, MECE options). Enable it from the /config output-style picker; it's opt-in and fixed at session start, so set it now or /clear after.

5. Recommend the Next Action

Branch on the derived state, first match wins, and say why:

• scope_state is not complete → the WBS doesn't exist yet; recommend consult-scope ("scope not done — let's frame the key question and derive the action fields").
• A field's state is unreadable → its manifest is broken, not unplanned; the surfaced warning is the recommendation — fix or inspect that field.json before any routing.
• A field has an empty deliverables[] (and is not unreadable) → the WBS has an unplanned container; recommend consult-action-fields to plan that field's deliverable set.
• Any deliverable carries lineage_status.status: "stale" → an upstream deliverable it depends on changed, so its artifact is out of date. Stale work outranks both in-progress and pending work here: finishing fresh work on a stale foundation wastes it, so refreshing comes first. Recommend refreshing the stale set in topological order — upstream before dependents: run deliverable-graph.py <engagement-dir> refresh-order and recommend the layer-0 deliverable(s) first (they depend on nothing else that is stale, so they are safe to refresh now); a deeper-layer deliverable is refreshed only once the layer above it has been. Route to knowledge-refresh for the research, then consult-design-thinking to re-run that deliverable's loop. Never recommend refreshing a dependent before its upstream dependency.
• scope_state is complete AND personas_gate is "pending" (no scope-seeded persona and no .gate-waiver marker yet) → the first design-thinking deliverable's empathize and test stages would run on degraded fallback, and the hard gate blocks a not-yet-started deliverable until personas are seeded. Recommend consult-personas to seed acting personas from the scope's Stakeholder dimension (or waive to defaults) before recommending any deliverable work. Placed after the scope/unreadable/stale checks so it never masks a broken manifest or a stale set, but ahead of the pending-deliverable recommendation so it front-runs the first deliverable. Read personas_gate from the engagement-status.sh rollup already fetched above; this stays a read-only route — never write persona state here. (This gate is naturally once-per-engagement: once satisfied it stays satisfied, and it never fires for an in-progress or resumed deliverable, which the branches below own.)
• A deliverable is in-progress → resume it where it stands; recommend consult-design-thinking naming the field, the deliverable, and its dt_stage ("competitor-map is mid-ideate — pick the loop back up there").
• A deliverable is complete but its persona_review is pending or in-progress → the acting-persona challenge hasn't closed; recommend consult-personas to run (or finish) the challenge pass.
• A deliverable is pending → start the next one; recommend consult-design-thinking (or the deliverable's own producing_route when it names a different skill).
• Everything is complete → say so — the engagement is complete by derivation — and offer consult-action-fields to extend the WBS if the consultant wants to add fields or deliverables. Per-deliverable publish/render next steps follow the publish offer below (read each complete deliverable's publish[] and offer consult-publish or the Claude Design handoff) — surface them only as elect-only offers, never as a standing menu item.

Four further offers surface only when the consultant's request or a deliverable's state calls for them — not as standing menu items:

• The consultant names an already-complete deliverable to revisit or modify (a rework request) → offer to reopen it and route to consult-design-thinking, naming the field, the deliverable, and the stage the rework should re-enter (often define or ideate). The reopen itself — the complete → in-progress Edit and the up-front cascade-stale of its downstream dependents — is owned by consult-design-thinking's Open-the-Loop step, so resume stays read-only; it routes, it does not write.
• A deliverable's stored chosen_framework is null (a legacy deliverable created before a framework was chosen) and the consultant wants to assign one → offer to set it inline rather than sending them on a separate consult-action-fields round-trip. "Inline" means the offer surfaces here in the recommendation flow; the actual field.json write is delegated to consult-action-fields (which owns the deliverable manifest), so resume's read-only contract holds. Surface this only when the framework gap is relevant to the next action — never as blanket nagging across every legacy deliverable.
• A deliverable is complete with a non-null chosen_framework whose conformance hasn't been verified → offer a framework-adherence review: dispatch the consult-framework-adherence-reviewer agent (engagement_dir, field_slug, deliverable_slug, plugin_root) to score the finished artifact against its stored framework's structure signature and report drift with concrete findings. This is a structural-conformance axis distinct from the persona-challenge (Test) pass, so it complements rather than duplicates consult-personas. The reviewer is read-only — it reports drift, it never rewrites the artifact — so resume stays read-only too; acting on a finding is a separate consult-design-thinking rework. Surface this only when the conformance question is relevant to the next action, not as a standing audit of every complete deliverable.
• A complete deliverable (its persona_review closed) is unpublished, or published but not yet rendered → offer the publish / render next step from its publish[] lineage, even before the whole engagement is complete: an empty/absent publish[] → offer /cogni-consult:consult-publish to produce a presentation-ready brief; a populated publish[] → name its brief_path(s) and point the consultant to hand them to Claude Design (claude.ai/design) to render. This is read-only over publish[] — consult-publish owns brief production; resume only routes. Surface it only when the consultant elects it or names that deliverable, never as a standing menu item.

Recommend one action, not a menu. On the consultant's confirmation, dispatch the named skill via Skill(...) with the engagement path as the in-session handoff (the target skills skip rediscovery on handoff).

Important Notes

• Read-only: this skill never writes consult-project.json, field.json, personas, or logs — state writes belong to the routed skills. See $CLAUDE_PLUGIN_ROOT/references/data-model.md for ownership.
• Derived, not stored: field and engagement completion are computed at read time by engagement-status.sh; never trust a stale summary over a fresh script run.
• One recommendation: the dashboard orients, the recommendation commits — a single next action with its reason beats a list of options.