replicas-engine 0.1.100 → 0.1.101

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/dist/src/index.js +23 -2
  2. package/package.json +1 -1
package/dist/src/index.js CHANGED
@@ -1091,7 +1091,7 @@ function parseReplicasConfigString(content, filename) {
1091
1091
  }
1092
1092
 
1093
1093
  // ../shared/src/engine/environment.ts
1094
- var DAYTONA_SNAPSHOT_ID = "13-04-2026-relay-codex-gating";
1094
+ var DAYTONA_SNAPSHOT_ID = "13-04-2026-royal-york";
1095
1095
 
1096
1096
  // ../shared/src/engine/types.ts
1097
1097
  var DEFAULT_CHAT_TITLES = {
@@ -3349,7 +3349,15 @@ Your primary resource is your context window. Every tool call result, every code
3349
3349
 
3350
3350
  ## Parallel execution
3351
3351
 
3352
- You can spawn multiple subagents in parallel by making multiple spawn_agent tool calls in a single response. Do this whenever you have independent tasks \u2014 do NOT call them sequentially if they don't depend on each other. For example, if you need to test three endpoints, spawn three subagents in one response, not one after another.
3352
+ You can spawn multiple subagents in parallel by making multiple spawn_agent tool calls in a single response. Do this whenever you have truly independent tasks \u2014 do NOT call them sequentially if they don't depend on each other. For example, if you need to test three endpoints, spawn three subagents in one response, not one after another.
3353
+
3354
+ **IMPORTANT \u2014 recognize implicit dependencies:** Tasks are NOT independent when one task's output shapes what the other needs to know. The most common case: if you are building both a backend and a frontend, the frontend depends on the backend \u2014 it needs to know the API endpoints, request/response shapes, data models, etc. You MUST complete the backend subagent first, then use its output to write an informed prompt for the frontend subagent. Other examples of implicit dependencies:
3355
+ - Building a library/SDK and code that consumes it
3356
+ - Creating a database schema and code that queries it
3357
+ - Implementing an API and writing tests for that API
3358
+ - Any producer/consumer relationship where the consumer needs to match the producer's interface
3359
+
3360
+ When in doubt about whether tasks are independent, run them sequentially. The cost of unnecessary sequencing is just latency; the cost of incorrect parallelization is a subagent working blind and producing incompatible code.
3353
3361
 
3354
3362
  ## Testing your work
3355
3363
 
@@ -3369,6 +3377,19 @@ The default subagent timeout is 10 minutes. For tasks you expect to take longer
3369
3377
 
3370
3378
  ${agentSelectionLines}
3371
3379
 
3380
+ ## Incremental progress on complex tasks
3381
+
3382
+ For complex, multi-step tasks (building a full feature, large refactors, new systems), do NOT try to accomplish everything in one shot. Instead, take an incremental commits approach \u2014 work step by step, feature by feature, commit by commit.
3383
+
3384
+ **How to work incrementally:**
3385
+ - Break the task into discrete, meaningful steps (e.g., schema first, then API endpoints, then frontend, then tests).
3386
+ - After each step, commit the work in a clean state before moving on. By clean state, we mean the kind of code that is stable, does not introduce major bugs, the code is orderly and well-documented. A good rule of thumb is: could another agent or human pick up from this commit and continue building without having to clean up an unrelated mess?
3387
+ - Review the results of each step before starting the next. This lets you catch issues early, adjust course, and carry accurate context forward about what was already built.
3388
+
3389
+ **Why this matters:** Trying to build an entire system in one shot tends to produce lower-quality, harder-to-debug output. Incremental commits create natural checkpoints \u2014 if something goes wrong, you only lose one step's worth of work, not everything. This approach also preserves your context window since you track high-level progress without holding all the implementation details.
3390
+
3391
+ **When to use this:** Apply incremental progress for tasks that span multiple components, require many files to be changed, or would otherwise be a large, complex, time-consuming piece of work. For simple, self-contained tasks (a bug fix, a single endpoint, a small refactor), working in one pass is fine.
3392
+
3372
3393
  ## Effective delegation
3373
3394
 
3374
3395
  - Subagents start with a blank context. Include all relevant file paths, code snippets, requirements, and constraints in the prompt.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "replicas-engine",
3
- "version": "0.1.100",
3
+ "version": "0.1.101",
4
4
  "description": "Lightweight API server for Replicas workspaces",
5
5
  "type": "module",
6
6
  "main": "dist/src/index.js",