dreamcontext 0.17.1 → 0.18.0

package/dist/skill-packs/agents/goal-implementer.md

@@ -31,17 +31,36 @@ skills:
   - dreamcontext
 ---
 
+## Forked from the planner — role re-binding (v2)
+
+You were **forked from the planner session** and inherit its full context — the goal,
+the codebase reads, the converged plan, the dependency-map table. That inheritance is
+free context, not a promotion. **You are NO LONGER the planner** — you are the
+**implementer for exactly ONE task/lane** in the dependency map.
+
+- Build **only the files your lane owns** (the `files owned` cell for your task in the
+  map). Never touch a file another lane owns, even if it looks related or convenient —
+  same-file-touching tasks are supposed to be the same lane; if you find yourself needing
+  to edit a file outside your lane, that's the plan being wrong (see Hard limits below),
+  not a reason to reach across lanes.
+- Treat every `depends on` lane's output as a **frozen, pinned contract** — exact
+  signatures/types as the plan stated them. Build against the contract, don't renegotiate it.
+- You may have full session context, but your **actions** are scoped to your one task.
+
 ## Skills always loaded
 
 - **engineering** — the implementation standard (security, error handling at
   boundaries, testing, idempotency, naming). Code that ignores it fails Phase 5 review.
+  This is **non-negotiable**: every implementer, on every lane, loads and follows it —
+  there is no fast path that skips it.
 - **dreamcontext** — the task at `_dream_context/state/<slug>.md` is your spec and
   source of truth; log progress with `dreamcontext tasks log <slug> "..."`.
 
 If the task touches a domain skill (`firebase-firestore`, `firebase-cloud-functions`,
 `claude-api`, etc.), load it before writing code.
 
-You are the **Goal Implementer**.
+You are the **Goal Implementer**. You think at **high** effort — enough to get your one
+lane right without the overhead the planner's xhigh pass already spent on the map.
 
 ## Mandate
 
@@ -56,10 +75,32 @@ Build **exactly** what the task's acceptance criteria require — no more, no le
 - On a re-implement after a FAIL, fix the **specific** failure reported by the
   reviewer/validator; don't churn unrelated code.
 
+## Wave/lane discipline
+
+- Honor your task's `files owned` and `depends on` cells in the dependency-map table —
+  they are the safety contract that makes parallel waves safe. Don't assume an
+  upstream lane's file is done until the wave's build+test gate has passed.
+- **Report back to the orchestrator; do not edit the dependency-map table yourself.**
+  The orchestrator is the map's **single writer** — implementers report progress and
+  findings, they don't write concurrent updates into the task doc's map.
+- Log your session id and progress via `dreamcontext tasks log <slug> "..."` so the
+  orchestrator can append you to the task's session registry — see "Session registry" in the goal-skill SKILL.md for the exact block shape; **do not write that block
+  yourself**, and do not reproduce it inline — cite it, let the orchestrator record it.
+
+## Running as a CLI builder session (v2)
+
+You are a **CLI builder session** (`claude -p`), forked from the planner via
+`--resume <plannerId> --fork-session`. A re-implement after a review/validation FAIL
+**resumes this same session** (`claude -p --resume <yourSessionId>`) — you already have
+full context, so fix only the **specific** reported failure; don't re-read files you
+already read or re-derive decisions already settled in this session.
+
 ## Hard limits
 
 - **Do not expand scope.** A nice-to-have you noticed is not in the criteria — note it,
   don't build it.
+- **Do not touch another lane's files.** If your task genuinely requires it, that's a
+  broken plan, not a judgment call — stop and report (below).
 - **If you discover the plan is wrong or impossible, STOP and report back to the
   orchestrator.** Do not silently redesign — the orchestrator may need to reopen Phase 1/2.
 - **Do not weaken a test to make it pass.** A failing test is signal, not an obstacle.

package/dist/skill-packs/agents/goal-plan-reviewer.md

@@ -38,6 +38,13 @@ skills:
 
 You are a **Goal Plan Reviewer**. You review a **plan**, before code exists.
 
+You are dispatched **clean and fresh every round**, fed only the plan text from the task
+doc — you never inherit the planner's session. That independence is what keeps your
+verdict trustworthy.
+
+Under the **critic** lens, you may also sanity-check the dependency-map table: are
+contracts pinned, are same-file tasks kept out of the same wave, are the waves acyclic?
+
 ## Your lens
 
 The orchestrator passes you ONE lens in the dispatch prompt. Review primarily through it:

package/dist/skill-packs/agents/goal-planner.md

@@ -47,6 +47,22 @@ shape the plan.
 
 You are the **Goal Planner**. Your output is a plan, not code.
 
+## Running as a CLI builder session (v2)
+
+You are spawned as a CLI session (`claude -p`), not a one-shot Agent-tool call.
+The orchestrator captures your `session_id` from the JSON output and records it in the task doc's session registry — see "Session registry" in the goal-skill SKILL.md. Every revision round after Phase 2 plan review **resumes
+this same session** (`--resume <your-session-id>`) with only the delta (the
+reviewers' new findings) — it does not re-send the whole goal. **Build on your
+prior turns; do not re-explore what you already read.** Each implementer forked
+from you in Phase 4 inherits your full context at cache-read price — keep your
+own exploration lean so their fork isn't taxed by dead ends.
+
+The orchestrator picks your model per tier via `--model` (opus for Large
+goals, sonnet for Medium/Small) when it spawns you; the `opus` in this file's
+frontmatter is only the fallback used if you're ever dispatched as a plain
+Agent-tool subagent instead of a CLI session. You think at **xhigh** — the
+dependency map must be right, since implementers build against it in parallel.
+
 ## Mandate
 
 Produce a plan an enthusiastic mid-level engineer could execute without guessing.
@@ -73,3 +89,21 @@ A structured markdown plan: Overview · File-by-file steps (CREATE/EDIT) · Acce
 criteria · Test plan · Risks & open questions · Out-of-scope. Honest and concrete. If a
 requirement is genuinely impossible or contradictory, say so plainly instead of
 inventing a path around it.
+
+For **Medium and Large** goals (the tier router decides), also include a
+**dependency-map table** with exactly these columns: `task | files owned |
+depends on | wave | contract`. Waves are computed from file ownership:
+
+- **Same file → same lane.** Two tasks that touch the same file are
+  auto-dependent — never place them in the same wave.
+- **Pin contracts.** For every dependency edge, state the exact signature/type
+  the upstream task freezes, so parallel tasks in later waves cannot diverge
+  from what they were promised.
+- The map is what lets the orchestrator dispatch implementers for a wave in
+  parallel (fork+resume) instead of serially — an imprecise map produces file
+  collisions or broken contracts downstream, so verify ownership against the
+  real files before finalizing it.
+- Small goals skip the map entirely; go straight to file-by-file steps.
+
+Do not write the session-registry block yourself — that literal template lives in
+the goal-skill SKILL.md and the orchestrator populates it as builders are spawned.

package/dist/skill-packs/catalog.json

@@ -164,7 +164,7 @@
     },
     {
       "name": "council",
-      "description": "Load when the user wants to debate a non-trivial decision, asks for a second opinion, says 'help me decide', 'run this by council', or invokes /council. Orchestrates 3\u201310 persona sub-agents across multiple rounds, then a synthesizer writes a final decision report.",
+      "description": "Load when the user wants to debate a non-trivial decision, asks for a second opinion, says 'help me decide', 'run this by council', or invokes /council. Orchestrates 3–10 persona sub-agents across multiple rounds, then a synthesizer writes a final decision report.",
       "tags": [
         "council",
         "debate",
@@ -223,7 +223,7 @@
     },
     {
       "name": "goal-skill",
-      "description": "Load when the user wants to drive a non-trivial goal to completion under sub-agent orchestration, or invokes /goal-skill. Triggers: 'drive this goal to completion', 'subagent-driven development', 'SDD', 'orchestrate this build', 'run this end-to-end with reviewers'. For multi-step features/fixes too large or risky for a single straight-line pass.",
+      "description": "Load when the user wants to drive a non-trivial goal to completion under sub-agent orchestration, or invokes /goal-skill. Triggers: 'drive this goal to completion', 'subagent-driven development', 'SDD', 'orchestrate this build', 'run this end-to-end with reviewers'. For multi-step features/fixes too large or risky for a single straight-line pass. v2: CLI builder sessions fork once and resume per round, clean fresh judges, a tier router (S/M/L) scales ceremony, and dependency-map waves parallelize implementation.",
       "tags": [
         "goal",
         "orchestration",
@@ -234,7 +234,7 @@
         "validation"
       ],
       "alwaysApply": false,
-      "base": "Subagent-driven goal completion: the main agent orchestrates plan -> plan-review -> task-doc -> implement -> code-review -> validate, iterating each gate (cap 3, then escalate) until validation passes against user-agreed criteria. Reuses the engineering reviewer for code review and the dreamcontext task system as the working doc.",
+      "base": "Subagent-driven goal completion: builders (planner, implementers) run as CLI sessions that fork once and resume per round, while judges (plan-reviewers, code reviewer, validator) stay clean and fresh each round, meeting the work only through the task doc. A tier router (S/M/L, with a hot-path override for auth/crypto/env/migrations) scales ceremony to goal size; a dependency-map table drives wave-parallel implementation; loops iterate to convergence by signal (new findings resume the builder, repeated findings trigger one re-fork, then escalate) rather than a fixed iteration count, until validation passes against user-agreed criteria. Reuses the engineering reviewer for code review and the dreamcontext task system as the working doc.",
       "subSkills": [],
       "relatedAgents": [
         "goal-planner",
@@ -245,6 +245,16 @@
       ],
       "crossPackDeps": [
         "engineering (planner/implementer/reviewer load engineering standards)"
+      ],
+      "assets": [
+        {
+          "file": "assets/goal-skill-viewer.cjs",
+          "dest": "goal-skill-viewer.cjs"
+        },
+        {
+          "file": "assets/goal-skill-demo.cjs",
+          "dest": "goal-skill-demo.cjs"
+        }
       ]
     }
   ],
@@ -338,7 +348,7 @@
         "marketing-monitor",
         "marketing-creative"
       ],
-      "corpus_status": "9 videos \u00b7 4 speakers \u00b7 draft \u00b7 agent roster: 2 active (strategy, monitor) + 1 stubbed (creative)"
+      "corpus_status": "9 videos · 4 speakers · draft · agent roster: 2 active (strategy, monitor) + 1 stubbed (creative)"
     },
     {
       "name": "system-prompts",
@@ -356,7 +366,7 @@
     {
       "name": "excalidraw",
       "file": "excalidraw/SKILL.md",
-      "description": "Generate or extend Obsidian Excalidraw (.excalidraw.md) boards — lay out images/screenshots, text, shapes, arrows, frames, lanes and grids — by writing a small JSON spec and running a deterministic script that emits valid plugin markup (near-zero tokens, always renders). Embeds local images via the plugin's sha1 wikilink trick (no base64). Triggers: 'draw this in excalidraw', 'make an excalidraw board', 'excalidraw diagram', 'funnel map board', 'embed screenshots into excalidraw', 'excalidraw çiz', 'excalidraw board oluştur'.",
+      "description": "Generate or extend Obsidian Excalidraw (.excalidraw.md) boards — charts (line/bar/compare/stacked/gantt/quadrant/donut/heatmap/table/timeline/KPI/sparkline), funnels, wireframes and real-proportion device mockups (iPhone/iPad/Mac), flowcharts, images/screenshots, shapes, arrows, frames, lanes and grids — by passing DATA to ~44 ready-made deterministic builders that emit valid plugin markup (near-zero tokens, always renders). Every builder works from JS or straight from a JSON spec; the build audits the scene for overlaps, buried labels and over-long lines. Embeds local images via the plugin's sha1 wikilink trick (no base64). Triggers: 'draw this in excalidraw', 'make an excalidraw board', 'excalidraw diagram', 'chart', 'gantt', 'funnel map board', 'wireframe', 'iphone mockup', 'embed screenshots into excalidraw', 'excalidraw çiz', 'excalidraw board oluştur', 'grafik çiz', 'wireframe çiz'.",
       "tags": [
         "excalidraw",
         "obsidian",
@@ -453,7 +463,7 @@
       "name": "marketing-strategy",
       "file": "agents/marketing-strategy.md",
       "pack": "meta-marketing",
-      "description": "Strategy Optimizer for Meta ads \u2014 plans cohort hypotheses (shape-validated), drafts campaign briefs grounded in the meta-marketing skill-pack corpus, always asks for budget, never launches.",
+      "description": "Strategy Optimizer for Meta ads — plans cohort hypotheses (shape-validated), drafts campaign briefs grounded in the meta-marketing skill-pack corpus, always asks for budget, never launches.",
       "tags": [
         "meta",
         "ads",
@@ -468,7 +478,7 @@
       "name": "marketing-monitor",
       "file": "agents/marketing-monitor.md",
       "pack": "meta-marketing",
-      "description": "Performance Monitor for live Meta campaigns \u2014 reads insights, applies post-launch rules from account-ops.md \u00a74, writes hypothesis-ledger entries to per-day learnings, recommends moves but never auto-mutates.",
+      "description": "Performance Monitor for live Meta campaigns — reads insights, applies post-launch rules from account-ops.md §4, writes hypothesis-ledger entries to per-day learnings, recommends moves but never auto-mutates.",
       "tags": [
         "meta",
         "ads",
@@ -482,7 +492,7 @@
       "name": "marketing-creative",
       "file": "agents/marketing-creative.md",
       "pack": "meta-marketing",
-      "description": "Creative Director for Meta ads \u2014 STUBBED in v0. Refuses all requests until marketing.creative_director.enabled = true. v1 will merge brainstorming + copy + asset generation.",
+      "description": "Creative Director for Meta ads — STUBBED in v0. Refuses all requests until marketing.creative_director.enabled = true. v1 will merge brainstorming + copy + asset generation.",
       "tags": [
         "meta",
         "ads",
@@ -509,7 +519,7 @@
       "name": "review-security",
       "file": "agents/review-security.md",
       "pack": "multi-review",
-      "description": "Security specialist in the multi-review team. Reviews ONLY for security defects \u2014 exploitable vulnerabilities, secret leakage, auth/authz gaps, injection, SSRF/CSRF/XSS, env-var exposure, weak crypto, insecure deserialization. Outputs a bounded greptile-style report.",
+      "description": "Security specialist in the multi-review team. Reviews ONLY for security defects — exploitable vulnerabilities, secret leakage, auth/authz gaps, injection, SSRF/CSRF/XSS, env-var exposure, weak crypto, insecure deserialization. Outputs a bounded greptile-style report.",
       "tags": [
         "review",
         "security",
@@ -523,7 +533,7 @@
       "name": "review-cloud-functions",
       "file": "agents/review-cloud-functions.md",
       "pack": "multi-review",
-      "description": "Cloud Functions specialist in the multi-review team. Reviews ONLY Firebase Cloud Functions and serverless function changes \u2014 idempotency, infinite-loop risk, scaling/concurrency, cold-start cost, retry semantics, secrets handling, unbounded fan-out, billing footguns, trigger-loop hazards.",
+      "description": "Cloud Functions specialist in the multi-review team. Reviews ONLY Firebase Cloud Functions and serverless function changes — idempotency, infinite-loop risk, scaling/concurrency, cold-start cost, retry semantics, secrets handling, unbounded fan-out, billing footguns, trigger-loop hazards.",
       "tags": [
         "review",
         "cloud-functions",
@@ -537,7 +547,7 @@
       "name": "review-frontend",
       "file": "agents/review-frontend.md",
       "pack": "multi-review",
-      "description": "Frontend specialist in the multi-review team. Reviews ONLY frontend changes (React/Vue/Svelte/Next/web components/CSS) \u2014 file-size scalability, hook correctness, accessibility, bundle bloat, render perf, state management, design-token discipline, i18n, and frontend-specific security (XSS sinks, token storage).",
+      "description": "Frontend specialist in the multi-review team. Reviews ONLY frontend changes (React/Vue/Svelte/Next/web components/CSS) — file-size scalability, hook correctness, accessibility, bundle bloat, render perf, state management, design-token discipline, i18n, and frontend-specific security (XSS sinks, token storage).",
       "tags": [
         "review",
         "frontend",
@@ -551,7 +561,7 @@
       "name": "review-edge-cases",
       "file": "agents/review-edge-cases.md",
       "pack": "multi-review",
-      "description": "Edge-case and production-risk specialist in the multi-review team. The paranoid one. Enumerates failure modes the other specialists don't own \u2014 empty/null/undefined inputs, off-by-one, concurrency, partial successes, retries, network failures, clock skew, timezone bugs, locale assumptions, state machine gaps, error swallowing. Always included for tier >= Lite.",
+      "description": "Edge-case and production-risk specialist in the multi-review team. The paranoid one. Enumerates failure modes the other specialists don't own — empty/null/undefined inputs, off-by-one, concurrency, partial successes, retries, network failures, clock skew, timezone bugs, locale assumptions, state machine gaps, error swallowing. Always included for tier >= Lite.",
       "tags": [
         "review",
         "edge-cases",
@@ -565,7 +575,7 @@
       "name": "biv-researcher",
       "file": "agents/biv-researcher.md",
       "pack": "business-idea-validation",
-      "description": "Research specialist for BIV pipeline. Executes Stage 1 (Problem Discovery \u2014 Reddit mining, forum analysis, Google Trends) and Stage 4 (Competitive Intelligence \u2014 competitor dossiers, gap analysis, 2x2 matrix). Heavy web research.",
+      "description": "Research specialist for BIV pipeline. Executes Stage 1 (Problem Discovery — Reddit mining, forum analysis, Google Trends) and Stage 4 (Competitive Intelligence — competitor dossiers, gap analysis, 2x2 matrix). Heavy web research.",
       "tags": [
         "biv",
         "research",
@@ -579,7 +589,7 @@
       "name": "biv-customer-analyst",
       "file": "agents/biv-customer-analyst.md",
       "pack": "business-idea-validation",
-      "description": "Customer analysis specialist for BIV pipeline. Executes Stage 2 (Customer Analysis \u2014 segmentation, personas, beachhead selection) and Stage 5 (AI-simulated Customer Validation from forum data).",
+      "description": "Customer analysis specialist for BIV pipeline. Executes Stage 2 (Customer Analysis — segmentation, personas, beachhead selection) and Stage 5 (AI-simulated Customer Validation from forum data).",
       "tags": [
         "biv",
         "customer",
@@ -593,7 +603,7 @@
       "name": "biv-market-analyst",
       "file": "agents/biv-market-analyst.md",
       "pack": "business-idea-validation",
-      "description": "Market sizing specialist for BIV pipeline. Executes Stage 3 (Market Sizing \u2014 TAM/SAM/SOM via top-down and bottom-up analysis, ambition-level gating).",
+      "description": "Market sizing specialist for BIV pipeline. Executes Stage 3 (Market Sizing — TAM/SAM/SOM via top-down and bottom-up analysis, ambition-level gating).",
       "tags": [
         "biv",
         "market",
@@ -608,7 +618,7 @@
       "name": "biv-strategist",
       "file": "agents/biv-strategist.md",
       "pack": "business-idea-validation",
-      "description": "Strategy specialist for BIV pipeline. Executes Stage 6 (Synthesis & Hypothesis \u2014 opportunity snapshot, riskiest assumptions, testable hypothesis, RICE features), Stage 7 (GTM Strategy \u2014 channels, messaging, first-100-users plan), and optional Lean Canvas capstone.",
+      "description": "Strategy specialist for BIV pipeline. Executes Stage 6 (Synthesis & Hypothesis — opportunity snapshot, riskiest assumptions, testable hypothesis, RICE features), Stage 7 (GTM Strategy — channels, messaging, first-100-users plan), and optional Lean Canvas capstone.",
       "tags": [
         "biv",
         "strategy",
@@ -622,7 +632,7 @@
       "name": "biv-financial-analyst",
       "file": "agents/biv-financial-analyst.md",
       "pack": "business-idea-validation",
-      "description": "Financial viability specialist for BIV pipeline. Executes Stage 8 (Year 1 P&L, unit economics \u2014 CAC/LTV/LTV:CAC, sensitivity analysis, Year 2 outlook).",
+      "description": "Financial viability specialist for BIV pipeline. Executes Stage 8 (Year 1 P&L, unit economics — CAC/LTV/LTV:CAC, sensitivity analysis, Year 2 outlook).",
       "tags": [
         "biv",
         "financial",
@@ -636,7 +646,7 @@
       "name": "biv-decision-gate",
       "file": "agents/biv-decision-gate.md",
       "pack": "business-idea-validation",
-      "description": "Final decision specialist for BIV pipeline. Executes Stage 9 (Go/No-Go \u2014 scores 8 dimensions across Desirability/Feasibility/Viability, applies risk penalties, produces verdict: GO/PIVOT/NO-GO with specific next steps).",
+      "description": "Final decision specialist for BIV pipeline. Executes Stage 9 (Go/No-Go — scores 8 dimensions across Desirability/Feasibility/Viability, applies risk penalties, produces verdict: GO/PIVOT/NO-GO with specific next steps).",
       "tags": [
         "biv",
         "decision",
@@ -649,7 +659,7 @@
       "name": "goal-planner",
       "file": "agents/goal-planner.md",
       "pack": "goal-skill",
-      "description": "Planning specialist for goal-skill. Produces a rigorous file-by-file implementation plan by reading the codebase + dreamcontext context. Does not write code or the task doc. Dispatched at Phase 1.",
+      "description": "Planning specialist for goal-skill. Runs as a CLI builder session (fork once, resume per revision round) and produces a rigorous file-by-file plan plus a dependency-map table (task/files owned/depends on/wave/contract) by reading the codebase + dreamcontext context, thinking xhigh. Does not write code or the task doc. Dispatched at Phase 1.",
       "tags": [
         "goal",
         "planning",
@@ -662,7 +672,7 @@
       "name": "goal-plan-reviewer",
       "file": "agents/goal-plan-reviewer.md",
       "pack": "goal-skill",
-      "description": "Plan reviewer for goal-skill. Critiques a prose plan (pre-code) from one lens (pragmatist/critic/security) and returns SOLID or NEEDS_WORK with blocking findings. Distinct from the code reviewer. Dispatched in parallel at Phase 2.",
+      "description": "Plan reviewer for goal-skill. Dispatched clean and fresh every round, fed only the plan text from the task doc (never the planner's session), and critiques it from one lens (pragmatist/critic/security), returning SOLID or NEEDS_WORK with blocking findings. Distinct from the code reviewer. Dispatched in parallel at Phase 2.",
       "tags": [
         "goal",
         "plan-review",
@@ -675,7 +685,7 @@
       "name": "goal-implementer",
       "file": "agents/goal-implementer.md",
       "pack": "goal-skill",
-      "description": "Implementation specialist for goal-skill. Builds strictly to a validated task's acceptance criteria, logs progress, does not expand scope, stops+reports if the plan is wrong. Dispatched at Phase 4.",
+      "description": "Implementation specialist for goal-skill. Forked from the planner's CLI session for exactly one task/lane of the dependency map, thinking high; builds strictly to that lane's owned files and the validated task's acceptance criteria, logs progress, does not expand scope or edit the dependency map, and stops+reports if the plan is wrong. Dispatched per wave at Phase 4 (fork once, resume per fix).",
       "tags": [
         "goal",
         "implementation",