@corbat-tech/coco 2.24.0 → 2.24.2

package/README.md

@@ -12,6 +12,9 @@
 
 </div>
 
+**Coco is an open-source CLI coding agent for real repositories.**  
+It plans work, edits files, runs tools/tests, and iterates until quality checks pass.
+
 ---
 
 ## What Is Coco?
@@ -20,6 +23,10 @@ Coco is a CLI coding agent for real projects. It can plan work, edit files, run
 
 Core idea: instead of a single "here is some code" response, Coco runs an implementation loop with validation and fixes.
 
+Best fit:
+- Teams and solo developers working on existing repos (not only greenfield demos).
+- Workflows that require multi-step execution and verification, not just text generation.
+
 ## What Coco Does
 
 - Multi-step execution in one run: explore -> implement -> test -> refine.
@@ -27,6 +34,8 @@ Core idea: instead of a single "here is some code" response, Coco runs an implem
 - Native tool use: files, git, shell, search/web, review, diff, build/test, MCP servers.
 - Multi-provider support (API, subscription, and local models).
 - Session-oriented REPL with slash commands, context compaction, and resumable workflows.
+- Reliability features for long sessions: provider retry/circuit-breaker, robust tool-call parsing, and safer stream error handling.
+- Replay harness support to reproduce agent-loop behaviors from fixtures for regression testing.
 
 Coco is designed to be useful on medium and large repos, not only toy examples.
 
@@ -81,6 +90,14 @@ On first run, Coco guides provider/model setup.
 
 Quality mode is configurable and can be turned on/off per session.
 
+## Reliability and Quality
+
+- Provider calls use retry and circuit-breaker protection by default (can be disabled with `COCO_PROVIDER_RESILIENCE=0`).
+- Tool-call handling is normalized across OpenAI/Codex-style streaming events to reduce malformed argument regressions.
+- Agent turns include quality telemetry (`score`, iteration usage, tool success/failure, repeated-output suppression).
+- Repeated identical tool outputs are suppressed in context to reduce token waste in multi-iteration loops.
+- Release readiness can be gated with `pnpm check:release` (typecheck + lint + stable provider/agent suites).
+
 ## Commands (REPL)
 
 Common commands:
@@ -182,6 +199,7 @@ pnpm install
 pnpm build
 pnpm test
 pnpm check
+pnpm check:release
 ```
 
 Tech stack:
@@ -192,12 +210,15 @@ Tech stack:
 - Zod
 - Commander
 
+Release gate (`pnpm check:release`) runs the stable typecheck/lint/provider+agent suites used for release readiness.
+
 ## Current Scope and Limitations
 
 - CLI-first product; VS Code extension source is in `vscode-extension/`.
 - Quality scores depend on project testability and model/tool quality.
 - Provider behavior can vary by endpoint/model generation.
 - Some advanced flows require external tooling (git, CI, MCP servers) to be installed/configured.
+- No agent can guarantee zero regressions; Coco is designed to reduce risk with verification loops, not to remove it entirely.
 
 ## Privacy