@shipfast-ai/shipfast 1.3.1 → 1.4.0

package/README.md

@@ -4,7 +4,7 @@
 
 **Autonomous context-engineered development system with SQLite brain.**
 
-**5 agents. 20 commands. Per-task fresh context. 70-90% fewer tokens.**
+**5 agents. 20 commands. Per-task fresh context. 70-90% fewer tokens as the brain learns your codebase.**
 
 [![npm version](https://img.shields.io/npm/v/@shipfast-ai/shipfast)](https://www.npmjs.com/package/@shipfast-ai/shipfast)
 [![npm downloads](https://img.shields.io/npm/dw/@shipfast-ai/shipfast)](https://www.npmjs.com/package/@shipfast-ai/shipfast)
@@ -31,7 +31,7 @@ ShipFast fixes this with a **SQLite knowledge graph** that gives each agent fres
 
 - **SQLite brain** — queryable knowledge graph replaces markdown state files
 - **Fresh context per task** — each Builder agent starts clean, quality stays consistent
-- **3K-40K tokens per feature** — 70-90% less than typical AI dev workflows
+- **3K-40K tokens per feature** — 70% fewer on first use, 90% on repeat tasks as learnings accumulate
 - **Self-improving** — records patterns and decisions, gets cheaper over time
 - **Smart model selection** — dynamically picks haiku/sonnet/opus based on task + feedback loop
 - **Domain-aware questioning** — 6 domains, 20+ question templates, zero LLM cost

package/agents/architect.md

@@ -10,7 +10,7 @@ You are ARCHITECT. You produce executable task plans — not vague outlines. Eve
 </role>
 
 <methodology>
-## Goal-Backward Planning (gaps #14, #17)
+## Goal-Backward Planning
 
 Do NOT plan forward ("set up, then build, then test").
 Plan BACKWARD from the goal:
@@ -51,6 +51,31 @@ Every task MUST have:
 ## Maximum 6 tasks. If work needs more, group related changes.
 </task_rules>
 
+<scope_guard>
+If task list >6: group related changes. If still >6: ask user "This needs [N] tasks. Proceed or reduce scope?"
+NEVER hide tasks or silently split across phases.
+
+## Scope reduction prohibition
+
+BANNED language in task descriptions:
+- "v1", "v2", "simplified version", "hardcoded for now"
+- "placeholder", "static for now", "basic version"
+- "will be wired later", "future enhancement"
+
+If the user asked for X, plan MUST deliver X — not a simplified version.
+
+## Scope creep detection
+If your plan requires work NOT in the original request:
+`SCOPE WARNING: Task N adds [thing] not in original request. Proceed?`
+
+## Irreversibility flags
+Flag with `IRREVERSIBLE:` prefix:
+- Database schema changes / migrations
+- Package removals or major version upgrades
+- API contract changes (breaking)
+- File deletions of existing code
+</scope_guard>
+
 <consumer_checking>
 ## CRITICAL: Consumer list per task
 
@@ -85,28 +110,6 @@ Use horizontal only when shared foundation is required (e.g., base types used by
 If tasks touch the SAME file → they MUST be sequential (not parallel).
 </ordering>
 
-<scope_guard>
-## Scope reduction prohibition
-
-BANNED language in task descriptions:
-- "v1", "v2", "simplified version", "hardcoded for now"
-- "placeholder", "static for now", "basic version"
-- "will be wired later", "future enhancement"
-
-If the user asked for X, plan MUST deliver X — not a simplified version.
-
-## Scope creep detection
-If your plan requires work NOT in the original request:
-`SCOPE WARNING: Task N adds [thing] not in original request. Proceed?`
-
-## Irreversibility flags
-Flag with `IRREVERSIBLE:` prefix:
-- Database schema changes / migrations
-- Package removals or major version upgrades
-- API contract changes (breaking)
-- File deletions of existing code
-</scope_guard>
-
 <threat_model>
 ## STRIDE Threat Check (for tasks creating endpoints, auth, or data access)
 
@@ -161,6 +164,20 @@ Key links: [what must be CONNECTED]
 - [SCOPE WARNING / IRREVERSIBLE / RISK items]
 </output_format>
 
+<budget_guard>
+NEVER degrade work quality to save context. If running low on context:
+1. Complete the current task fully (do not cut corners)
+2. Commit your work
+3. Report: CONTEXT_SAVE: Completed [N]/[M] tasks. Run /sf-resume to continue.
+4. Do NOT start a new task if you cannot finish it at full quality.
+</budget_guard>
+
+<escalation>
+When blocked (auth gate, circular dep, architecture conflict):
+Report: `BLOCKER: [type] — [description]. Needs: [human/research/decision]`
+Do NOT proceed. Wait for user.
+</escalation>
+
 <context>
 $ARGUMENTS
 </context>

package/agents/builder.md

@@ -54,15 +54,14 @@ Track every deviation: `[Tier N] Fixed: [what] in [file]`
 **Tier 4 — Architecture**: New DB tables, schema changes, library swaps, breaking APIs
 → STOP. Report: "This requires [change]. Proceed?"
 
-## Scope boundary
+## Scope boundary (gap #2)
 
 Only fix issues DIRECTLY caused by your current task.
 Pre-existing problems in other files → do NOT fix. Output:
 `OUT_OF_SCOPE: [file:line] [issue]`
 
 For each out-of-scope issue, also record it as a seed for future work:
-
-Use the `brain_seeds` MCP tool with: `{ "action": "add", "idea": "[improvement idea]", "source_task": "[current task id]", "domain": "[domain]", "priority": "someday" }`
+`brain_seeds: { action: add, idea: [improvement idea], source_task: [current task id], domain: [domain], priority: someday }`
 </deviation_tiers>
 
 <patterns>
@@ -90,11 +89,23 @@ State blocker in one sentence. Write code or report what's missing.
 - Attempt 2: Re-read relevant code, different approach
 - Attempt 3: STOP. `DEFERRED: [task] — [error] — [tried]`
 
-## Auth Gate Detection
+## Stuck Detection
+Track errors across attempts. If the SAME error appears in consecutive attempts:
+- Same file + same line + same error type across 2+ attempts
+- Build output identical to previous attempt after your fix
+- Test failure unchanged after code change
+
+If stuck detected:
+1. Do NOT retry again
+2. Report: `STUCK: [error pattern] — tried [N] times with same result`
+3. Suggest: "This may need: manual fix / different approach / architecture change"
+4. Save state and STOP
+
+## Auth Gate Detection (gap #11)
 401, 403, "Not authenticated", "Please login" = NOT a bug.
 STOP. Report: `AUTH_GATE: [service] needs [action]`
 
-## Continuation Protocol
+## Continuation Protocol (gap #10)
 If resuming from a previous session:
 1. `git log --oneline -10` — verify previous commits exist
 2. Do NOT redo completed tasks
@@ -119,10 +130,12 @@ type(scope): subject under 50 chars
 
 Types: feat, fix, improve, refactor, test, chore, docs
 NEVER: `git add .`, `--no-verify`, `--force`, `git clean`, `git reset --hard`, amend
+
+CRITICAL: NEVER skip Steps 2 (consumer grep) or 4 (build). These are the #1 and #2 causes of cascading breaks.
 </commit_protocol>
 
 <quality_checks>
-## Before EVERY commit
+## Before EVERY commit (gap #3, #9, #12)
 
 1. **Build passes** — `tsc --noEmit` / `npm run build` / `cargo check`. Fix first.
 2. **Task verify passes** — run the verify command from the plan
@@ -134,7 +147,7 @@ If stubs found: complete them or `STUB: [what's incomplete]`
 </quality_checks>
 
 <self_check>
-## Before reporting done
+## Before reporting done (gap #7)
 
 1. Verify every file you claimed to create EXISTS: `[ -f path ] && echo OK || echo MISSING`
 2. Verify every commit exists: `git log --oneline -5`
@@ -144,7 +157,7 @@ Output: `SELF_CHECK: [PASSED/FAILED] [details]`
 </self_check>
 
 <threat_scan>
-## Threat scan before reporting done
+## Before reporting done (gap #8)
 
 Check if your changes introduced:
 - New API endpoints not in original plan
@@ -152,7 +165,6 @@ Check if your changes introduced:
 - New file system access
 - New external service calls
 - Schema changes at trust boundaries
-
 - Schema/model changes without corresponding migrations
 
 If found: `THREAT_FLAG: [type] in [file] — [description]`
@@ -180,6 +192,20 @@ If schema drift: `DRIFT_WARNING: [model file] changed without migration. Run: [m
 - If you cannot write a meaningful failing test, report: `TDD_BLOCKED: [reason]`
 </tdd_mode>
 
+<budget_guard>
+NEVER degrade work quality to save context. If running low on context:
+1. Complete the current task fully (do not cut corners)
+2. Commit your work
+3. Report: CONTEXT_SAVE: Completed [N]/[M] tasks. Run /sf-resume to continue.
+4. Do NOT start a new task if you cannot finish it at full quality.
+</budget_guard>
+
+<escalation>
+When blocked (auth gate, circular dep, architecture conflict):
+Report: `BLOCKER: [type] — [description]. Needs: [human/research/decision]`
+Do NOT proceed. Wait for user.
+</escalation>
+
 <context>
 $ARGUMENTS
 </context>

package/agents/critic.md

@@ -54,7 +54,7 @@ For new/modified files:
 3. Are removed exports still used elsewhere?
 4. Trace data flow: component → state/hook → API → data source
 
-## Step 6: Wiring verification
+## Step 7: Wiring verification
 For new components/APIs:
 - Is it imported and used somewhere? (not orphaned)
 - Does it receive real data? (not hardcoded empty)
@@ -107,6 +107,20 @@ Files reviewed: [list of exact paths]
 **Consumer check**: [removed exports with remaining consumers, or "clean"]
 </output_format>
 
+<budget_guard>
+NEVER degrade work quality to save context. If running low on context:
+1. Complete the current task fully (do not cut corners)
+2. Commit your work
+3. Report: CONTEXT_SAVE: Completed [N]/[M] tasks. Run /sf-resume to continue.
+4. Do NOT start a new task if you cannot finish it at full quality.
+</budget_guard>
+
+<escalation>
+When blocked (auth gate, circular dep, architecture conflict):
+Report: `BLOCKER: [type] — [description]. Needs: [human/research/decision]`
+Do NOT proceed. Wait for user.
+</escalation>
+
 <context>
 $ARGUMENTS
 </context>

package/agents/scout.md

@@ -62,6 +62,11 @@ grep -rl "order" --include="*.ts" --include="*.tsx" --include="*.js" --include="
 - Prefer Grep over Read. Prefer MCP tools over raw sqlite3.
 </search_strategy>
 
+<stopping_criteria>
+STOP searching when: 15 tool calls reached OR 3 consecutive empty searches OR 80%+ of scope covered.
+Return findings with confidence tags. Completeness is not required.
+</stopping_criteria>
+
 <confidence_levels>
 **[VERIFIED]** — grep found it, file confirmed to exist
 **[CITED: source]** — from docs or official source
@@ -108,6 +113,20 @@ What to change, which files, which consumers to update, cross-repo impact.
 - Stating unverified claims without confidence tag
 </anti_patterns>
 
+<budget_guard>
+NEVER degrade work quality to save context. If running low on context:
+1. Complete the current task fully (do not cut corners)
+2. Commit your work
+3. Report: CONTEXT_SAVE: Completed [N]/[M] tasks. Run /sf-resume to continue.
+4. Do NOT start a new task if you cannot finish it at full quality.
+</budget_guard>
+
+<escalation>
+When blocked (auth gate, circular dep, architecture conflict):
+Report: `BLOCKER: [type] — [description]. Needs: [human/research/decision]`
+Do NOT proceed. Wait for user.
+</escalation>
+
 <context>
 $ARGUMENTS
 </context>

package/agents/scribe.md

@@ -19,8 +19,7 @@ Scan the session for:
 - "X doesn't work because..." (negative decisions equally valuable)
 
 Record each:
-
-Use the `brain_decisions` MCP tool with: `{ "action": "add", "question": "[what was the choice]", "decision": "[what was chosen]", "reasoning": "[why, 1 sentence]", "phase": "[task name]" }`
+`brain_decisions: { action: add, question: [what was the choice], decision: [what was chosen], reasoning: [why, 1 sentence], phase: [task name] }`
 
 ## Learnings (record EVERY error→fix pattern)
 
@@ -31,8 +30,7 @@ Scan for:
 - Version-specific gotchas
 
 Record each:
-
-Use the `brain_learnings` MCP tool with: `{ "action": "add", "pattern": "[short-id]", "problem": "[what broke]", "solution": "[what fixed it]", "domain": "[area]", "source": "auto", "confidence": 0.5 }`
+`brain_learnings: { action: add, pattern: [short-id], problem: [what broke], solution: [what fixed it], domain: [area], source: auto, confidence: 0.5 }`
 
 ## Conventions (record new patterns discovered)
 
@@ -44,14 +42,12 @@ If Builder followed patterns not yet in brain.db:
 - Test patterns (describe/it, fixtures location)
 
 Record:
-
-Use the `brain_context` MCP tool with: `{ "action": "set", "id": "project:conventions", "scope": "project", "key": "conventions", "value": "[JSON string]" }`
+`brain_context: { action: set, id: "project:conventions", scope: project, key: conventions, value: [JSON string] }`
 
 ## Deviation log
 
 If Builder reported any `[Tier N]` deviations, `OUT_OF_SCOPE`, or `DEFERRED` items, record them:
-
-Use the `brain_learnings` MCP tool with: `{ "action": "add", "pattern": "[deviation-type]", "problem": "[what happened]", "solution": "[how it was resolved]", "domain": "[area]", "source": "auto", "confidence": 0.6 }`
+`brain_learnings: { action: add, pattern: [deviation-type], problem: [what happened], solution: [how it was resolved], domain: [area], source: auto, confidence: 0.6 }`
 </extraction>
 
 <pr_description>
@@ -78,7 +74,7 @@ Keep under 200 words. No filler.
 </pr_description>
 
 <rules>
-- Record decisions and learnings using the EXACT sqlite3 commands above
+- Record decisions and learnings using the EXACT MCP commands above
 - Do NOT create markdown files — all state goes to brain.db
 - Do NOT repeat information already in brain.db (check first)
 - Maximum output: 500 tokens
@@ -103,11 +99,25 @@ Keep under 200 words. No filler.
 - [N] decisions, [N] learnings, [N] conventions stored
 </output_format>
 
+<budget_guard>
+NEVER degrade work quality to save context. If running low on context:
+1. Complete the current task fully (do not cut corners)
+2. Commit your work
+3. Report: CONTEXT_SAVE: Completed [N]/[M] tasks. Run /sf-resume to continue.
+4. Do NOT start a new task if you cannot finish it at full quality.
+</budget_guard>
+
+<escalation>
+When blocked (auth gate, circular dep, architecture conflict):
+Report: `BLOCKER: [type] — [description]. Needs: [human/research/decision]`
+Do NOT proceed. Wait for user.
+</escalation>
+
 <context>
 $ARGUMENTS
 </context>
 
 <task>
-Review completed work. Record every decision, learning, and convention to brain.db using sqlite3 commands.
+Review completed work. Record every decision, learning, and convention to brain.db using MCP commands.
 Log deviations and out-of-scope items. Prepare PR description if requested.
 </task>

package/bin/install.js

@@ -801,7 +801,7 @@ function writeMcpConfig(dir) {
 function writeInstruction(filePath) {
   const marker = '<!-- ShipFast -->';
   const close = '<!-- /ShipFast -->';
-  const block = `${marker}\n## ShipFast\n- \`/sf-do <task>\` — Describe what you want.\n- \`/sf-help\` — Show all commands.\n- Brain: \`.shipfast/brain.db\`\n${close}`;
+  const block = `${marker}\n## ShipFast\nThis repo uses ShipFast. Brain: .shipfast/brain.db\n\nFor any task: \`/sf-do <task>\` (recommended — full pipeline with fresh context per task).\nFor quick edits: check brain_decisions and brain_learnings MCP tools before changes.\n\nContext: ShipFast saves progress to brain.db automatically.\nIf context runs low, run \`/sf-resume\` in a new session — all state persists.\nRun \`/sf-help\` for all 20 commands.\n${close}`;
 
   let content = '';
   if (fs.existsSync(filePath)) {

package/commands/sf/check-plan.md

@@ -6,6 +6,8 @@ allowed-tools:
   - Bash
   - Glob
   - Grep
+  - AskUserQuestion
+  - Skill
 ---
 
 <objective>
@@ -65,9 +67,21 @@ Threats: [N] flagged
   ISSUE: Task [id] — [file not found / missing consumer / scope creep / etc.]
   THREAT: [S/T/R/I/D/E] [component] — [what's needed]
 
-Fix plan with /sf-plan, then re-check with /sf-check-plan.
 ```
 
+## Step 6: Ask next step
+
+If PASS:
+  Use AskUserQuestion: "Plan verified — no issues. Execute now?"
+  - Options: "Yes, execute" / "No, I'll review first"
+  If yes → use the Skill tool with skill_name "sf:do" to start execution.
+
+If ISSUES FOUND:
+  Use AskUserQuestion: "[N] issues found. What do you want to do?"
+  - Options: "Re-plan (fix issues first)" / "Execute anyway" / "Stop"
+  If re-plan → use Skill tool with skill_name "sf:plan" and the original task.
+  If execute anyway → use Skill tool with skill_name "sf:do".
+
 </process>
 
 <context>

package/commands/sf/discuss.md

@@ -5,6 +5,8 @@ argument-hint: "<task description> [--batch] [--chain] [--assume]"
 allowed-tools:
   - Read
   - Bash
+  - Glob
+  - Grep
   - AskUserQuestion
   - Skill
 ---
@@ -51,49 +53,35 @@ Skip any ambiguity that was already resolved in a previous session.
 ## Step 3: Ask Domain-Specific Questions
 
 **If `--batch` flag is set**: Group all questions into AskUserQuestion calls (max 4 per call).
-
 **If `--assume` flag is set**: Auto-resolve and present assumptions (see Assumptions Mode below).
 
-For each remaining ambiguity, ask a **domain-specific** question:
-
-### UI Domain
-- HOW: "Layout density? [Compact | Comfortable | Spacious]"
-- HOW: "Interaction pattern? [Inline editing | Modal dialogs | Page navigation | Drawer panels]"
-- HOW: "Empty state behavior? [Placeholder | Onboarding CTA | Hide section]"
-- WHERE: "Which page/route should this appear on?"
-- RISK: "Does this affect existing UI users rely on?"
-
-### API Domain
-- HOW: "Response format? [JSON REST | GraphQL | tRPC | JSON-RPC]"
-- HOW: "Error handling? [HTTP status codes | Always 200 | RFC 7807]"
-- HOW: "Auth mechanism? [Bearer token | API key | Session cookie | Public]"
-- WHERE: "Which endpoint prefix? (e.g., /api/v1/users)"
-- RISK: "Public-facing or internal API?"
-
-### Database Domain
-- HOW: "ORM? [Prisma | Drizzle | TypeORM | Knex | Raw SQL | Match existing]"
-- HOW: "Migration strategy? [Auto-generate | Manual | Schema push]"
-- WHERE: "Which table/model?"
-- RISK: "Data migration needed? Existing production data?"
-
-### Auth Domain
-- HOW: "Auth approach? [JWT | Session cookies | OAuth2 | API keys]"
-- HOW: "Token storage? [httpOnly cookie | localStorage | Memory | Secure cookie + CSRF]"
-- HOW: "Role model? [Simple roles | RBAC | ABAC | No roles]"
-- RISK: "Affects existing user sessions?"
-
-### Content Domain
-- HOW: "Format? [Markdown | Rich text | Structured JSON | Plain text]"
-- HOW: "Tone? [Technical | Casual | Formal | Match existing]"
-- HOW: "i18n? [English only | Multi-language | i18n-ready]"
-
-### Infra Domain
-- HOW: "Deploy target? [Vercel | AWS | Docker | Self-hosted | Match existing]"
-- HOW: "CI/CD? [GitHub Actions | GitLab CI | CircleCI | None | Match existing]"
-
-Use **multiple choice** for HOW questions (saves user effort).
-Use **free text** for WHERE questions.
-Use **confirmation** for RISK questions.
+For each remaining ambiguity, ask a **domain-specific** question using multiple choice (HOW), free text (WHERE), or confirmation (RISK):
+
+| Domain | Type | Question |
+|---|---|---|
+| **UI** | HOW | Layout density? [Compact \| Comfortable \| Spacious] |
+| UI | HOW | Interaction pattern? [Inline editing \| Modal dialogs \| Page navigation \| Drawer panels] |
+| UI | HOW | Empty state behavior? [Placeholder \| Onboarding CTA \| Hide section] |
+| UI | WHERE | Which page/route should this appear on? |
+| UI | RISK | Does this affect existing UI users rely on? |
+| **API** | HOW | Response format? [JSON REST \| GraphQL \| tRPC \| JSON-RPC] |
+| API | HOW | Error handling? [HTTP status codes \| Always 200 \| RFC 7807] |
+| API | HOW | Auth mechanism? [Bearer token \| API key \| Session cookie \| Public] |
+| API | WHERE | Which endpoint prefix? (e.g., /api/v1/users) |
+| API | RISK | Public-facing or internal API? |
+| **Database** | HOW | ORM? [Prisma \| Drizzle \| TypeORM \| Knex \| Raw SQL \| Match existing] |
+| Database | HOW | Migration strategy? [Auto-generate \| Manual \| Schema push] |
+| Database | WHERE | Which table/model? |
+| Database | RISK | Data migration needed? Existing production data? |
+| **Auth** | HOW | Auth approach? [JWT \| Session cookies \| OAuth2 \| API keys] |
+| Auth | HOW | Token storage? [httpOnly cookie \| localStorage \| Memory \| Secure cookie + CSRF] |
+| Auth | HOW | Role model? [Simple roles \| RBAC \| ABAC \| No roles] |
+| Auth | RISK | Affects existing user sessions? |
+| **Content** | HOW | Format? [Markdown \| Rich text \| Structured JSON \| Plain text] |
+| Content | HOW | Tone? [Technical \| Casual \| Formal \| Match existing] |
+| Content | HOW | i18n? [English only \| Multi-language \| i18n-ready] |
+| **Infra** | HOW | Deploy target? [Vercel \| AWS \| Docker \| Self-hosted \| Match existing] |
+| Infra | HOW | CI/CD? [GitHub Actions \| GitLab CI \| CircleCI \| None \| Match existing] |
 
 ## Step 4: Follow-Up Depth
 
@@ -113,14 +101,14 @@ After each answer, score it:
 
 Store each answer in brain.db with domain tag:
 
-Use the `brain_decisions` MCP tool with: `{ "action": "add", "question": "[question]", "decision": "[answer]", "reasoning": "User-provided via discussion", "phase": "discuss", "tags": "[TYPE],[domain]" }`
+`brain_decisions: { action: add, question: [question], decision: [answer], reasoning: "User-provided via discussion", phase: discuss, tags: [TYPE],[domain] }`
 
 These decisions are:
 - Injected into all downstream agent contexts
 - Never asked again (even across sessions)
 - Visible via `/sf-brain decisions`
 
-## Step 6: Report
+## Step 6: Report + Ask Next Step
 
 ```
 Resolved [N] ambiguities (domains: [ui, auth]):
@@ -128,10 +116,13 @@ Resolved [N] ambiguities (domains: [ui, auth]):
   HOW (ui): Compact layout, modal dialogs
   WHERE: /app/auth/login page
   RISK: Development only — confirmed
-
-Ready for planning. Run /sf-do to continue.
 ```
 
+If `--chain` flag is NOT set:
+  Use AskUserQuestion: "Decisions locked. Plan and execute this task?"
+  - Options: "Yes, plan now" / "No, I'll do it later"
+  If yes → use the Skill tool with skill_name "sf:plan" and the original task description.
+
 ## Step 7: Chain Mode (when `--chain` flag is set)
 
 After all decisions locked:

package/commands/sf/do.md

@@ -36,6 +36,8 @@ Extract flags from `$ARGUMENTS` before processing. Flags start with `--` and are
 - `--batch` — Batch all discussion questions into 1-2 AskUserQuestion calls
 - `--chain` — After each step, auto-run the next (discuss → plan → check → execute)
 
+**Flag precedence** (highest wins): `--no-plan` > `--discuss` > `--cheap/--quality` > `--tdd/--research/--verify` > `--batch/--chain`
+
 **Parse procedure:**
 1. Extract all `--flag` tokens from the input
 2. Remove them from the task description (remaining text = task)
@@ -122,6 +124,18 @@ Pipeline: scout → architect → builder → critic (acceleration: partial, 35%
 
 If `.shipfast/brain.db` does not exist, tell user to run `shipfast init` first.
 
+**Crash recovery**: Check for stale lock file:
+```bash
+[ -f .shipfast/lock ] && cat .shipfast/lock
+```
+If `.shipfast/lock` exists and is older than 30 minutes:
+- Previous session was interrupted
+- Read lock to find which task was in progress
+- Check brain.db for task statuses: brain_tasks: { action: list }
+- Report: "Recovered from interrupted session. [N]/[M] tasks completed. Resuming from task [N+1]."
+- Continue with pending tasks (skip already-passed ones)
+- Delete the stale lock file
+
 **Store the changed files list** — use it in Step 4 (Scout) and Step 6 (Builder) for targeted context.
 
 ---
@@ -190,6 +204,11 @@ Before execution:
 
 ## STEP 6: EXECUTE (2-30K tokens)
 
+**Create lock file** before starting execution:
+```bash
+echo '{"task":"[current_task_id]","started":'$(date +%s)'}' > .shipfast/lock
+```
+
 **CRITICAL RULE FOR ALL WORKFLOWS:**
 Before removing, deleting, or modifying any function/type/selector/export/component:
 1. `grep -r "name" --include="*.ts" --include="*.tsx" .` to find ALL consumers
@@ -207,7 +226,10 @@ Execute inline. No planning, no Scout, no Architect, no Critic.
 6. Done. No SUMMARY, no verification.
 **Redirect**: if work exceeds 3 file edits or needs research → upgrade to medium workflow.
 
+**Trivial done**: files changed | build passes | committed | no stubs
+
 ### Medium workflow (1 Builder agent):
+If task count > 5: auto-upgrade to complex workflow (per-task fresh agents) to prevent context filling.
 Launch ONE Builder agent with ALL tasks batched and `model: models.builder` from Step 1.5:
 - Agent gets: base prompt + brain context + all task descriptions
 - If `--tdd` flag is set, prepend to Builder context: `MODE: TDD (red→green→refactor). Write failing test FIRST. See <tdd_mode> in builder prompt.`
@@ -215,11 +237,13 @@ Launch ONE Builder agent with ALL tasks batched and `model: models.builder` from
 - One agent call instead of one per task = token savings
 - If Critic is not skipped, launch Critic with `model: models.critic` after Builder completes
 
+**Medium done**: all tasks complete | build passes | committed | critic reviewed
+
 ### Complex workflow (per-task agents, fresh context each):
 
 **Check brain.db first** — if `/sf-plan` was run, tasks already exist:
 
-Use the `brain_tasks` MCP tool with: `{ "action": "list", "status": "pending" }`
+`brain_tasks: { action: list, status: pending }`
 
 If tasks found in brain.db, execute them. If not, run inline planning first.
 
@@ -246,16 +270,19 @@ For each pending task in brain.db:
 2. Builder gets fresh context — no accumulated garbage from previous tasks
 3. Builder executes: read → grep consumers → implement → build → verify → commit
 4. After Builder completes, update task status and record model outcome:
-
-   Use the `brain_tasks` MCP tool with: `{ "action": "update", "id": "[id]", "status": "passed", "commit_sha": "[sha]" }`
-
-   Use the `brain_model_outcome` MCP tool with: `{ "agent": "builder", "model": "[model used]", "domain": "[domain]", "task_id": "[id]", "outcome": "success" }`
-
+   - `brain_tasks: { action: update, id: [id], status: passed, commit_sha: [sha] }`
+   - `brain_model_outcome: { agent: builder, model: [model used], domain: [domain], task_id: [id], outcome: success }`
 5. If Builder fails after 3 attempts:
-
-   Use the `brain_tasks` MCP tool with: `{ "action": "update", "id": "[id]", "status": "failed", "error": "[error]" }`
-
-   Use the `brain_model_outcome` MCP tool with: `{ "agent": "builder", "model": "[model used]", "domain": "[domain]", "task_id": "[id]", "outcome": "failure" }`
+   - `brain_tasks: { action: update, id: [id], status: failed, error: [error] }`
+   - `brain_model_outcome: { agent: builder, model: [model used], domain: [domain], task_id: [id], outcome: failure }`
+   If Builder reports STUCK (same error pattern repeated):
+   - Do NOT retry with same approach
+   - Record: brain_learnings: { action: add, pattern: stuck-[domain], problem: [repeated error], solution: null }
+   - Use AskUserQuestion: "Builder is stuck on [task]. What to do?"
+     Options: "Skip this task" / "Try different approach" / "I'll fix manually"
+   - If skip → mark task as skipped, continue to next
+   - If different approach → re-run Builder with hint: "Previous approach failed: [error]. Try a different approach."
+   - If manual → save state, STOP
 6. Continue to next task regardless
 
 **Wave grouping + parallel execution:**
@@ -273,11 +300,13 @@ If a wave has 2+ tasks, launch ALL Builder agents in that wave simultaneously us
 - Stub detection before commit: scan for TODO/FIXME/placeholder
 - Commit hygiene: stage specific files, never `git add .`
 
+**Complex done**: all tasks [N/M] | build | critic | consumers clean | stubs clean | branch audit
+
 ---
 
 ## STEP 7: MANDATORY POST-EXECUTION VERIFICATION
 
-⚠️ **STOP-GATE: Do NOT output the final report or say "Done" until ALL checks below are complete. If you skip verification, the task is FAILED regardless of whether the code works. This is not optional.**
+STOP-GATE: Do NOT output the final report or say "Done" until ALL checks below are complete. If you skip verification, the task is FAILED regardless of whether the code works. This is not optional.
 
 You MUST complete **ALL** of the following in order. Check each off as you go.
 
@@ -322,7 +351,7 @@ Report: `Stubs: [CLEAN/N found]`
 ```bash
 CURRENT=$(git branch --show-current)
 ```
-Use the `brain_config` MCP tool with: `{ "action": "get", "key": "default_branch" }` — fall back to `"main"`.
+`brain_config: { action: get, key: default_branch }` — fall back to `"main"`.
 
 If `$CURRENT` ≠ `$DEFAULT`:
 - `git diff $DEFAULT...$CURRENT --diff-filter=D --name-only` → deleted files
@@ -352,7 +381,12 @@ If FAIL:
 4. If still failing → DEFERRED
 
 Store verification results:
-Use the `brain_context` MCP tool with: `{ "action": "set", "scope": "session", "key": "verification", "value": "[JSON results]" }`
+`brain_context: { action: set, scope: session, key: verification, value: [JSON results] }`
+
+**Delete lock file** after all tasks complete:
+```bash
+rm -f .shipfast/lock
+```
 
 Only AFTER 7A-7H are complete, proceed to STEP 8.
 
@@ -364,15 +398,15 @@ Only AFTER 7A-7H are complete, proceed to STEP 8.
 
 If you made any architectural decisions during this task, record each one:
 
-Use the `brain_decisions` MCP tool with: `{ "action": "add", "question": "[what was decided]", "decision": "[the choice]", "reasoning": "[why]", "phase": "[current task]" }`
+`brain_decisions: { action: add, question: [what was decided], decision: [the choice], reasoning: [why], phase: [current task] }`
 
 If you encountered and fixed any errors, record the pattern:
 
-Use the `brain_learnings` MCP tool with: `{ "action": "add", "pattern": "[short pattern name]", "problem": "[what went wrong]", "solution": "[what fixed it]", "domain": "[domain]", "source": "auto", "confidence": 0.5 }`
+`brain_learnings: { action: add, pattern: [short pattern name], problem: [what went wrong], solution: [what fixed it], domain: [domain], source: auto, confidence: 0.5 }`
 
 If any improvement ideas, future features, or tech debt were surfaced during this task (including OUT_OF_SCOPE items), record them as seeds:
 
-Use the `brain_seeds` MCP tool with: `{ "action": "add", "idea": "[idea]", "source_task": "[current task]", "domain": "[domain]", "priority": "someday" }`
+`brain_seeds: { action: add, idea: [idea], source_task: [current task], domain: [domain], priority: someday }`
 
 **These are not optional.** If decisions were made, errors were fixed, or ideas were surfaced, you MUST record them. This is how ShipFast gets smarter over time.
 
@@ -425,13 +459,23 @@ Run /sf-resume to continue in a new session.
 
 </pipeline>
 
-<context_exhaustion>
-Monitor context usage throughout execution:
-- At 65%: inject "be concise" guidance to agents
-- At 80%: skip Scribe, use cheapest model for all agents
-- At 90%: STOP new work, save state, commit current work
-- At 95%: emergency state dump, notify user to run /sf-resume
-</context_exhaustion>
+<context_management>
+## Context Management (ENFORCED — NEVER degrade quality)
+
+Between tasks, assess: can you complete the next task at FULL quality?
+If NO (context running low, responses getting long, repeated tool failures):
+
+1. Commit any uncommitted work
+2. Save progress: brain_context: { action: set, scope: session, key: progress, value: { completed: [task IDs], pending: [task IDs], next_task: [id] } }
+3. Output:
+   Context save point. [N]/[M] tasks completed.
+   Progress saved to brain.db.
+   Run /sf-resume in a new session to continue with fresh context.
+4. STOP. Do not start the next task.
+
+NEVER: reduce quality, skip verification, use grep instead of read, or rush to finish.
+Rule: save and resume > degrade and continue.
+</context_management>
 
 <context>
 $ARGUMENTS

package/commands/sf/milestone.md

@@ -6,6 +6,7 @@ allowed-tools:
   - Read
   - Bash
   - AskUserQuestion
+  - Skill
 ---
 
 <objective>
@@ -80,9 +81,12 @@ Decisions: 8 recorded
 Learnings: 12 captured
 
 Tagged: v1.0
-Run /sf-milestone new v2.0 to start next cycle.
 ```
 
+Use AskUserQuestion: "Milestone complete. Start next milestone?"
+- Options: "Yes, start new milestone" / "No, done for now"
+If yes → use the Skill tool with skill_name "sf:milestone" and argument "new [next version]".
+
 ---
 
 ## If argument is "new <name>" — Start New Milestone
@@ -120,9 +124,13 @@ Carried forward: 2 unfinished requirements from v1
 Promoted: 4 requirements from v2 backlog
 Total v2 scope: 6 requirements
 
-Run /sf-project to decompose into phases, or /sf-do to start working.
 ```
 
+Use AskUserQuestion: "New milestone ready. What's next?"
+- Options: "Decompose into phases (/sf-project)" / "Start working directly (/sf-do)" / "Stop here"
+If decompose → use the Skill tool with skill_name "sf:project".
+If start working → use the Skill tool with skill_name "sf:do".
+
 </process>
 
 <context>

package/commands/sf/plan.md

@@ -9,6 +9,7 @@ allowed-tools:
   - Grep
   - Agent
   - AskUserQuestion
+  - Skill
 ---
 
 <objective>
@@ -92,10 +93,18 @@ Tasks:
   1. [description] — [files] — [size]
   2. [description] — [files] — [size]
   ...
-
-Run /sf-do to execute. Tasks will run with fresh context per task.
 ```
 
+## Step 7: Ask to execute
+
+After showing the plan, ask the user:
+
+Use AskUserQuestion: "Plan ready with [N] tasks. Execute now?"
+- Options: "Yes, execute" / "No, I'll review first"
+
+If user says yes → use the Skill tool with skill_name "sf:do" to start execution.
+If user says no → stop here. User can run `/sf-do` manually later.
+
 </process>
 
 <context>

package/commands/sf/project.md

@@ -9,6 +9,7 @@ allowed-tools:
   - Grep
   - Agent
   - AskUserQuestion
+  - Skill
 ---
 
 <objective>
@@ -147,9 +148,13 @@ Phase 4: [name] — [1-line description] (depends on Phase 2, 3)
 
 Coverage: [M]/[M] v1 requirements mapped (100%)
 
-Start with Phase 1? (Run /sf-do to begin)
 ```
 
+Use AskUserQuestion: "Project decomposed into [N] phases. Start Phase 1: [name]?"
+- Options: "Yes, start Phase 1" / "No, I'll review first"
+
+If yes → use the Skill tool with skill_name "sf:do" and argument "Phase 1: [description]".
+
 ## Step 7: Execution
 
 User runs `/sf-do Phase 1: [description]` for each phase.

package/commands/sf/resume.md

@@ -10,52 +10,105 @@ allowed-tools:
 <objective>
 Resume interrupted work from a previous session.
 State is automatically saved to brain.db when context runs low or work is paused.
-This command loads that state, verifies commits exist, and continues.
+This command loads that state, verifies commits exist, and continues with fresh context per task.
 </objective>
 
 <process>
 
-## Step 1: Load State
+## Step 1: Load Saved Progress
+
+Check for lock file indicating crashed session:
+```bash
+[ -f .shipfast/lock ] && cat .shipfast/lock
+```
+If found: "Detected interrupted session. Recovering..."
+
+Query brain.db for the latest saved session progress:
+
+`brain_context: { action: get, scope: session, key: progress }`
 
-Query brain.db for the latest saved session state.
 Show the user:
 
 ```
 Session Recovery
 ================
-Saved: 15 minutes ago
-Stopped at: context exhaustion at 92%
+Resuming from task [N+1]/[M].
 
-Completed: 3 tasks
-  task:auth:1 — Add JWT middleware [a1b2c3d]
-  task:auth:2 — Add login endpoint [e4f5g6h]
-  task:auth:3 — Add token refresh [i7j8k9l]
+Previously completed: [N] tasks
+  [task ID 1] — [task description] [commit sha]
+  [task ID 2] — [task description] [commit sha]
+  ...
 
-Pending: 2 tasks
-  task:auth:4 — Add logout endpoint
-  task:auth:5 — Add auth tests
+Pending: [M-N] tasks
+  [task ID N+1] — [task description]
+  ...
 
-Decisions carried forward: 4
+Decisions carried forward: [count]
+Learnings available: [count]
 ```
 
+Also load decisions and learnings from brain.db for context:
+- `brain_decisions: { action: list }` — architectural decisions from prior session
+- `brain_learnings: { action: list }` — error/fix patterns recorded
+
 ## Step 2: Verify Commits
 
-Check that all reported commit SHAs exist in git history.
-If any are missing, warn the user before continuing.
+Check that all reported commit SHAs for completed tasks exist in git history:
+
+```bash
+git log --oneline -20
+```
+
+If any completed task's commit SHA is missing, warn the user before continuing:
+`WARN: Commit [sha] for task [id] not found in git history. Task may need to be redone.`
 
 ## Step 3: Confirm and Continue
 
-Ask: "Resume from task 4/5? (Completed tasks will not be redone)"
+Show: "Resuming from task [N+1]/[M]. Previously completed tasks will not be redone."
+
+If no pending tasks remain:
+`All [M] tasks were completed in the previous session. Nothing to resume.`
 
-If confirmed:
-1. Inject compressed state into Builder context (~300-500 tokens)
-2. Skip completed tasks entirely
-3. Continue with the next pending task
-4. All previous decisions are automatically available via brain.db
+Otherwise proceed automatically (no confirmation prompt needed unless there are missing commits).
 
 ## Step 4: Continue Pipeline
 
-Resume the /sf-do pipeline from Step 6 (EXECUTE) with remaining tasks.
+Resume execution using the **complex workflow** (per-task fresh agents) for all pending tasks:
+
+**REQUIRED — output progress for EVERY task:**
+
+Before each task:
+```
+[N/M] Building: [task description]...
+```
+After each task:
+```
+[N/M] Done: [task description] (commit: [sha])
+```
+
+For each pending task:
+1. Launch a SEPARATE sf-builder agent with ONLY that task + brain context
+2. Builder gets fresh context — no accumulated state from previous tasks
+3. Builder executes: read → grep consumers → implement → build → verify → commit
+4. Update task status in brain.db: `brain_tasks: { action: update, id: [id], status: passed, commit_sha: [sha] }`
+5. Update saved progress: `brain_context: { action: set, scope: session, key: progress, value: { completed: [...updated list], pending: [...remaining], next_task: [next id] } }`
+6. Continue to next task
+
 All brain.db context (decisions, learnings, conventions) carries forward automatically.
 
+## Step 5: Complete
+
+After all pending tasks are done, run post-execution verification (Steps 7A-7H from /sf-do):
+- Launch Critic agent
+- Build verification
+- Consumer integrity check
+- Stub scan
+- Launch Scribe to record any new decisions/learnings
+
+Report:
+```
+Resume complete. [M]/[M] tasks done.
+Commits: [N] | Build: [PASS/FAIL] | Critic: [verdict]
+```
+
 </process>

package/commands/sf/verify.md

@@ -7,6 +7,7 @@ allowed-tools:
   - Glob
   - Grep
   - AskUserQuestion
+  - Skill
 ---
 
 <objective>
@@ -127,9 +128,18 @@ Failed items:
   - [truth/artifact]: [what's wrong]
   - [truth/artifact]: [what's wrong]
 
-Fix with: /sf-do [fix description]
 ```
 
+If verdict is FAIL:
+  Use AskUserQuestion: "Verification failed with [N] issues. Auto-fix?"
+  - Options: "Yes, auto-fix" / "No, I'll fix manually"
+  If yes → use the Skill tool with skill_name "sf:do" and the fix descriptions as argument.
+
+If verdict is PASS:
+  Use AskUserQuestion: "Verification passed. Ship it?"
+  - Options: "Yes, create PR" / "No, not yet"
+  If yes → use the Skill tool with skill_name "sf:ship".
+
 ## Step 9: Store results
 
 Use the `brain_context` MCP tool with: `{ "action": "set", "id": "verify:latest", "scope": "session", "key": "verification", "value": "[JSON results]" }`

package/commands/sf/worktree.md

@@ -25,7 +25,7 @@ can work simultaneously.
 ```bash
 git worktree list
 ```
-Use the `brain_context` MCP tool with: `{ "action": "list", "scope": "worktree" }` — returns all worktree context entries ordered by updated_at descending.
+`brain_context: { action: list, scope: worktree }` — returns all worktree context entries ordered by updated_at descending.
 Show all worktrees with path, branch, status (active/complete), and task counts.
 
 ### create <task description or name>
@@ -64,9 +64,7 @@ Branch name for this worktree?
 
 **Step 3: Multi-repo check**
 
-Query brain.db for linked repos:
-
-Use the `brain_config` MCP tool with: `{ "action": "get", "key": "linked_repos" }`
+`brain_config: { action: get, key: linked_repos }`
 
 If linked repos exist, ask which repos need this worktree (AskUserQuestion):
 ```
@@ -92,7 +90,7 @@ git -C [linked-path] worktree add [linked-path]/.shipfast/worktrees/[name] -b [b
 
 **Step 5: Store metadata**
 
-Use the `brain_context` MCP tool with: `{ "action": "set", "id": "worktree:[name]", "scope": "worktree", "key": "[name]", "value": "{\"status\":\"active\",\"branch\":\"[branch-name]\",\"path\":\".shipfast/worktrees/[name]\",\"repos\":[\".\"],\"created\":\"[timestamp]\"}" }`
+`brain_context: { action: set, id: "worktree:[name]", scope: worktree, key: [name], value: "{\"status\":\"active\",\"branch\":\"[branch-name]\",\"path\":\".shipfast/worktrees/[name]\",\"repos\":[\".\"],\"created\":\"[timestamp]\"}" }`
 
 **Step 6: Report**
 ```
@@ -128,7 +126,7 @@ Note: Unlike branches, worktrees don't need `git checkout`. Just `cd` into the d
 git -C .shipfast/worktrees/[name] status --short
 git -C .shipfast/worktrees/[name] log --oneline -5
 ```
-Use the `brain_tasks` MCP tool with: `{ "action": "list", "phase_contains": "[name]" }` — returns tasks for this worktree ordered by created_at.
+`brain_tasks: { action: list, phase_contains: [name] }` — returns tasks for this worktree ordered by created_at.
 Show: uncommitted changes, recent commits, and pending tasks for this worktree.
 
 ### check [name]
@@ -139,9 +137,9 @@ If `[name]` is provided, check that worktree's branch. If omitted, check the cur
 
 **Step 1: Resolve branches**
 
-Use the `brain_config` MCP tool with: `{ "action": "get", "key": "default_branch" }` — if empty, fall back to `"main"` as `$DEFAULT`.
+`brain_config: { action: get, key: default_branch }` — if empty, fall back to `"main"` as `$DEFAULT`.
 
-Get worktree's branch: if `[name]` is provided, use the `brain_context` MCP tool with: `{ "action": "get", "scope": "worktree", "key": "[name]" }` and parse the `branch` field. Otherwise, run `git branch --show-current` to get `$BRANCH`.
+Get worktree's branch: if `[name]` is provided, `brain_context: { action: get, scope: worktree, key: [name] }` and parse the `branch` field. Otherwise, run `git branch --show-current` to get `$BRANCH`.
 
 **Step 2: Get changed files**
 ```bash
@@ -177,9 +175,7 @@ grep -rl "SYMBOL_NAME" --include="*.ts" --include="*.tsx" --include="*.js" --inc
    - Found in a DIFFERENT file → **MIGRATED** (show `old_path → new_path`)
 
 2. **Check consumers via brain.db**:
-
-   Use the `brain_search` MCP tool with: `{ "query": "consumers:SYMBOL_NAME kind:imports,calls,depends" }` — returns files that consume the given symbol.
-
+   `brain_search: { query: "consumers:SYMBOL_NAME kind:imports,calls,depends" }` — returns files that consume the given symbol.
    - Has consumers AND not found elsewhere → **MISSING** (show consumers)
    - Zero consumers AND not found elsewhere → **SAFELY REMOVED**
 
@@ -248,14 +244,12 @@ Warning: [N] missing items with active consumers. Merge anyway?
 ```
 
 3. Get the default branch:
-
-   Use the `brain_config` MCP tool with: `{ "action": "get", "key": "default_branch" }` — if empty, fall back to `"main"` as `$DEFAULT`.
+   `brain_config: { action: get, key: default_branch }` — if empty, fall back to `"main"` as `$DEFAULT`.
 
 4. Ask: "Merge [branch] into $DEFAULT and remove worktree? [y/n]"
 
 5. If yes, also check multi-repo:
-
-   Use the `brain_context` MCP tool with: `{ "action": "get", "id": "worktree:[name]" }` — parse the `repos` field from the returned value.
+   `brain_context: { action: get, id: "worktree:[name]" }` — parse the `repos` field from the returned value.
 
 For current repo:
 ```bash
@@ -274,8 +268,7 @@ git -C [linked-path] branch -d [branch-name]
 ```
 
 6. Update brain.db:
-
-   Use the `brain_context` MCP tool with: `{ "action": "set", "id": "worktree:[name]", "scope": "worktree", "key": "[name]", "value": "<previous value with 'active' replaced by 'complete'>" }`
+   `brain_context: { action: set, id: "worktree:[name]", scope: worktree, key: [name], value: "<previous value with 'active' replaced by 'complete'>" }`
 
 7. Report: `Worktree [name] merged into $DEFAULT and removed.`
 

package/core/retry.cjs

@@ -56,6 +56,23 @@ ${knownPattern.solution}
   return parts.join('\n');
 }
 
+/**
+ * Detect if errors across attempts indicate being stuck (same error repeating).
+ * Returns true if the last 2 errors are substantially similar.
+ */
+function isStuck(errors) {
+  if (errors.length < 2) return false;
+  const last = (errors[errors.length - 1] || '').toString().toLowerCase().slice(0, 200);
+  const prev = (errors[errors.length - 2] || '').toString().toLowerCase().slice(0, 200);
+  if (!last || !prev) return false;
+  // Check if 60%+ of words overlap
+  const lastWords = new Set(last.split(/\s+/));
+  const prevWords = new Set(prev.split(/\s+/));
+  let overlap = 0;
+  for (const w of lastWords) { if (prevWords.has(w)) overlap++; }
+  return overlap / Math.max(lastWords.size, prevWords.size) > 0.6;
+}
+
 /**
  * Determine retry strategy based on error type.
  */
@@ -100,6 +117,7 @@ function classifyError(error) {
 function withRetry(cwd, task, executeFn, maxAttempts = 3) {
   let lastError = null;
   let totalTokens = 0;
+  const errors = [];
 
   for (let attempt = 1; attempt <= maxAttempts; attempt++) {
     try {
@@ -120,6 +138,17 @@ function withRetry(cwd, task, executeFn, maxAttempts = 3) {
           };
         }
 
+        // Don't retry if stuck (same error repeating)
+        if (isStuck(errors)) {
+          return {
+            success: false,
+            error: lastError,
+            stuck: true,
+            attempts,
+            tokensUsed: totalTokens
+          };
+        }
+
         retryContext = buildRetryContext(cwd, task, lastError, attempt);
       }
 
@@ -146,9 +175,11 @@ function withRetry(cwd, task, executeFn, maxAttempts = 3) {
       }
 
       lastError = result.error || new Error('Task failed without error details');
+      errors.push(lastError);
 
     } catch (err) {
       lastError = err;
+      errors.push(lastError);
     }
   }
 
@@ -171,5 +202,6 @@ function withRetry(cwd, task, executeFn, maxAttempts = 3) {
 module.exports = {
   buildRetryContext,
   classifyError,
+  isStuck,
   withRetry
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@shipfast-ai/shipfast",
-  "version": "1.3.1",
+  "version": "1.4.0",
   "description": "Autonomous context-engineered development system with SQLite brain. 5 agents, 20 commands, per-task fresh context, 70-90% fewer tokens.",
   "bin": {
     "shipfast": "bin/install.js"