Skip to content
CAI
Software that uses CAICheck a score

affaan-m/ECC

51.7

Adequate · 13 November 2026

180.2k

lines of production code

JavaScript

with Rust, Python

8

measurements over time

CAI band scale
CAI trend line
CAI lens gauges

What this system is

Features

Add CodeBuddy IDE support with install and uninstall scripts

Users can now install ECC workflows (commands, agents, skills, and rules) into the CodeBuddy IDE. This location provides the installation scripts (install.js, install.sh) and uninstall scripts (uninstall.js, uninstall.sh) that manage the lifecycle of these files within a CodeBuddy project, including manifest tracking and safe removal.

.codebuddy · high confidence

Add Kiro IDE formatting and quality gate scripts

New shell scripts have been added to the .kiro/scripts directory to support the Kiro IDE integration. The format.sh script automatically detects and runs code formatting using either Biome or Prettier based on project configuration files. The quality-gate.sh script performs a comprehensive project health check by sequentially running build, type checking, linting, and test suites, detecting the appropriate package manager and tooling (such as pnpm, yarn, npm, Biome, ESLint, Ruff, or golangci-lint) to provide a unified pass/fail status for project quality.

.kiro/scripts · high confidence

Add eval-harness example demonstrating static inspection and offline verification

The eval-harness example now provides a complete offline workflow for evaluating code variants without executing them. It demonstrates that candidate execution is refused without a verified OS backend, performs static inspection to detect syntactic warnings (such as hidden network calls or checker probes), replays locally declared fixtures, and builds/verifies offline capsule receipts. The example includes a taskset, baseline, candidate, and reward-hack variant fixtures to illustrate these capabilities.

examples/eval-harness · high confidence

Add hardened CLI test harness for Docker plugin setup

Introduces a new Docker-based test harness in \docker/plugin-setup\ to validate the ECC plugin installation process in isolated environments. The setup includes a hardened Dockerfile building on \node:22-bookworm-slim\ with \claude-code\ version 2.1.220, and a \compose.yaml\ defining services for fixture tests and real CLI execution (including a networked mode). The harness features a \codex-auth-relay.js\ to bridge OAuth callbacks, scripts to prepare and validate packed CLI packages (\prepare-packed-cli.js\, \verify-install-plan.js\), and utilities to resolve project directories and run platform-specific tests (\run-platform-tests.js\). This enables verification of install plans, target confinement, and CLI behavior in a controlled, read-only container environment.

docker/plugin-setup · high confidence

Add read-only coordination inventory example with declared goals and sessions support

The \examples/coordination-inventory\ directory now includes a new read-only inventory tool that generates local JSON reports by joining declared task IDs, parent IDs, heartbeat age, OS RAM, resource leases, and import warnings. This update extends the inventory's manifest contract (v1) to support optional \goals\ and \sessions\ collections, allowing users to observe declared intent (active/complete goals, open/closed sessions) alongside task data. The example includes synthetic fixtures (\manifest.json\, \goals.json\) and scripts (\evaluate.js\, \benchmark.js\) to characterize overlap detection accuracy and performance, demonstrating how declared goals and sessions are tracked without triggering automatic coordination actions.

examples/coordination-inventory · high confidence

Added evaluation case stubs and reference implementations for complex scenarios

The repository now includes a comprehensive set of new evaluation cases under \docker/context-profiles/complex-eval\, providing both 'naive' (intentionally flawed) and 'reference' (correct) implementations for various system design challenges. These new files cover incident triage (with adjusted timeout and logging logic), a Sentinel paste API (with security improvements like HTML escaping and path traversal protection), webhook relays (with retry logic), event statistics APIs (with optimized indexing), and idempotent payment webhooks. Additionally, stubs for a CLI tool and a Keccak selector are provided to support further evaluation work.

(repo-wide) · high confidence

Added soft-limit overage scenario for gateguard effectiveness testing

A new test scenario named 'soft-limit-overage' has been added to the gateguard effectiveness suite. This scenario evaluates how well the system handles a policy where usage overages are permitted and billed later, with a specific exception for the 'free' plan which refuses requests once usage exceeds 110% of its limit. The entry includes a reference implementation of the \checkQuota\ function, a grader to verify correct behavior against three specific test cases, and an intent definition that highlights the ambiguity between hard and soft limits to test the model's ability to adhere to the specified policy.

docker/gateguard-effectiveness/scenarios-intent/soft-limit-overage · high confidence

Cursor-specific ECC memory isolation and hook configuration

This change introduces scaffold files for Cursor that isolate ECC memory from Claude Code. A new \ecc-agent-data.json\ configures the memory root to \\~/.cursor/ecc\ via the \agentDataHome\ field, and a \hooks.json\ file registers a \sessionStart\ hook to execute \node .cursor/hooks/session-start.js\. Additionally, a rule file (\ecc-agent-data-home.mdc\) documents this isolated persistence strategy, ensuring session summaries and learned skills are stored separately from \\~/.claude\ unless explicitly shared.

scaffolds · high confidence

Cursor-to-Claude Code hook adapter and lifecycle hooks

This change introduces a new adapter layer in .cursor/hooks that translates Cursor's stdin JSON payloads into the format expected by Claude Code hooks, enabling seamless integration between the two environments. The adapter includes lifecycle hooks for session start/end, file edits, shell execution, and MCP calls, with built-in safety features like input size limits, sensitive file warnings, and secret detection in prompts. Users benefit from enhanced security through fail-closed behavior on oversized inputs and blocking of sensitive file reads, while maintaining backward compatibility with existing Claude Code hook infrastructure.

.cursor/hooks · high confidence

Initial repository scaffolding and configuration baseline

Establishes the foundational project structure and configuration files, including the MIT license, code of conduct, and agent instruction guides (AGENTS.md, CLAUDE.md). Adds developer tooling configurations for Prettier, ESLint, markdownlint, and CodeRabbit, alongside environment variable examples and package manager settings. Introduces the initial changelog documenting versions 1.10.0 through 2.2.3, and configures the default MCP server for Chrome DevTools.

(repo-wide) · high confidence

Introduce .trae directory with install and uninstall scripts for Trae IDE

Adds a new .trae directory containing install.sh and uninstall.sh scripts, along with documentation, to deploy Everything Claude Code (ECC) workflows into Trae projects. Users can now install commands, agents, skills, and rules locally or globally (to the home directory) using the install script, which supports a CN environment via the TRAE\_ENV variable. The uninstall script safely removes only ECC-managed files tracked in a manifest, preserving user-added content.

.trae · high confidence

Introduce ECC 2.0 agentic IDE control plane with session management and inter-agent communication

This change scaffolds the ECC 2.0 core in \ecc2/src\, establishing a new agentic IDE control plane. It introduces a structured inter-agent messaging system (\comms\) supporting task handoffs, queries, and conflict detection with priority levels. The new CLI (\main\) provides commands to start, delegate, and assign agent sessions, launch orchestration templates, and manage worktree policies. It also includes a bounded harness evaluation framework (\harness\_eval\) for deterministic configuration testing and a notification system (\notifications\) for desktop alerts, webhooks, and quiet hours.

ecc2/src · high confidence

Introduce ECC 2.0 session management and observability infrastructure

This change introduces the core ECC 2.0 session management layer, including a background daemon for session lifecycle, heartbeat monitoring, and automatic recovery of crashed sessions. It adds a new observability module that tracks tool calls, computes risk scores based on file sensitivity and blast radius, and suggests actions (Allow, Review, Require Confirmation, Block). The session manager now supports creating sessions with profiles and grouping, handling remote dispatch, and managing worktree auto-merge and auto-prune. A new output store buffers and broadcasts session output lines for real-time dashboard streaming. The harness detection system is expanded to support multiple IDE agents (Claude, Codex, Gemini, Cursor, etc.) and detects them via project markers. State persistence is handled via a new SQLite-based state store with comprehensive session, worktree, and daemon activity tracking.

ecc2/src/session · high confidence

Introduce ECC 2.0 terminal UI dashboard

The ECC 2.0 terminal UI is now available, providing a comprehensive dashboard for managing agent sessions, worktrees, and team coordination. Users can monitor session states, view real-time output, and manage budgets via token and cost meters with configurable alert thresholds. The interface supports interactive pane navigation, split-diff viewing for worktree changes, and direct controls for session lifecycle actions like spawning, stopping, resuming, and deleting. It also includes features for auto-dispatch, worktree merging, and conflict resolution, all accessible through a configurable keybinding system.

ecc2/src/tui · high confidence

Introduce ECC 2.0 worktree management and Git integration layer

This change introduces the core worktree management module for the ECC 2.0 agentic IDE control plane. It adds the ability to create and remove isolated Git worktrees for agent sessions, enforcing configurable branch prefixes and sharing dependency caches across worktrees. The module also provides a comprehensive Git status and patch view, exposing merge readiness checks, conflict previews, and structured data for staged/unstaged changes, file previews, and hunk-level actions to support the new UI controls and diff viewer.

ecc2/src/worktree · high confidence

Introduce ECC for Kiro: installable agents, skills, and steering files

This change adds the \.kiro/\ directory containing the ECC (Enhanced Claude Code) integration for Kiro. It includes an \install.sh\ script that non-destructively copies 33 agents, 43 skills, 22 steering files, 13 IDE hooks, and example settings into a Kiro project. The agents provide specialized workflows for code review, testing, and build resolution across multiple languages (TypeScript, Python, Go, Rust, Java, etc.), while the skills offer on-demand workflows like TDD and security reviews. The steering files define always-on rules for coding style, security, and testing, with language-specific patterns loaded via file matching. An example \mcp.json\ is also provided to configure Model Context Protocol servers.

.kiro · high confidence

Introduce ECC plugin hooks for OpenCode integration

Added a new plugin system in .opencode/plugins that bridges ECC hooks to OpenCode's event model, enabling features like auto-formatting on file edits and tracking changed files. The implementation includes a new ecc-hooks.ts file that maps OpenCode events (such as file.edited) to ECC actions, with resilience improvements to prevent session crashes if optional libraries are missing, and supports configurable hook profiles (minimal, standard, strict) via environment variables.

.opencode/plugins · high confidence

Introduce LLM abstraction layer with provider-agnostic interface and CLI selector

This change introduces a new LLM abstraction layer (\src/llm\) that provides a unified interface for interacting with multiple AI backends. It includes a core interface (\LLMProvider\) and specific adapters for OpenAI, Anthropic Claude, Ollama, Atlas Cloud, and Astraflow. The package also adds a CLI tool (\llm-select\) that allows users to interactively choose a provider and model, saving the configuration to a \.llm.env\ file. Additionally, it provides a prompt builder to normalize messages and tool definitions across different provider formats, and reasoning-stripping helpers for self-hosted models.

src/llm · high confidence

Introduce OpenCode plugin support with migration guide and configuration

The .opencode directory now provides a complete integration for the OpenCode CLI, including an opencode.json configuration file that defines agents, commands, and skills, alongside a TypeScript plugin module (index.ts) that exports hooks and custom tools. A new MIGRATION.md guide helps users transition from Claude Code by mapping hooks to OpenCode's plugin events, and a .npmignore file ensures clean package distribution. This change adds the necessary configuration and plugin infrastructure to run ECC features within the OpenCode environment.

.opencode · high confidence

Introduce Pi Coding Agent adapter for ECC integration

Adds a new \.pi\ directory containing a thin adapter that connects the Pi coding agent to ECC's canonical skills, commands, and engineering rules. The adapter mounts ECC's skills and commands directly without duplication, injects portable engineering rules into the system prompt, and maps Pi's session lifecycle events to ECC's existing hook runner (honoring \ECC\_HOOK\_PROFILE\ and \ECC\_DISABLED\_HOOKS\). It includes a \/ecc-doctor\ diagnostic command to verify the integration and handles hook execution safely via \execFile\ with isolated timeouts and output limits.

.pi · high confidence

Introduce agent self-evaluation and automated behavioral compliance skills

This change adds two new capabilities to the platform: an agent self-evaluation skill and a behavioral compliance measurement skill. The self-evaluation skill provides a standalone script that scores agent outputs against a five-axis rubric (Accuracy, Completeness, Clarity, etc.) using keyword heuristics and structural checks. The compliance skill introduces a grader that validates observation traces against YAML-defined compliance specs, handling temporal ordering constraints (before/after steps) and redacting sensitive operator home paths from reports to ensure privacy. Also in this area: 1 commit (1 fix) fixing .agents, llm-abstraction; 4 commits (4 fixes) fixing llm-abstraction.

llm-abstraction · medium confidence · unverified

Introduce layered TOML configuration system for ECC 2.0

The ECC 2.0 control plane now supports a structured, layered configuration system loaded from TOML files. This change introduces a comprehensive set of configuration modules that allow users to define and customize agent profiles (including model selection, tool permissions, and budget limits), orchestration templates for multi-step workflows, and various memory connectors (supporting JSONL and Markdown formats for files and directories, as well as Dotenv integration). Additionally, the configuration layer enables fine-grained control over session behaviors, including configurable budget thresholds with alert levels, risk scoring thresholds for tool actions, conflict resolution strategies, worktree management policies, and notification settings for desktop, webhook, and completion summaries.

ecc2/src/config · high confidence

Native 'orch-review' workflow pilot for autonomous code review

This change introduces a pilot native Claude Code workflow (\orch-review.workflow.js\) that ports the autonomous, fan-out-heavy review segment of the existing orchestration pipeline to the native engine. The workflow runs multi-dimensional code reviews (quality, language-specific, and conditional security) in parallel, deduplicates findings, and adversarially verifies all CRITICAL and HIGH issues before returning a verdict. It is invoked via the \/orch-review\ command, which gathers diffs and passes them to the workflow, ensuring that the review process benefits from barrier-free pipelining, automatic concurrency capping, and structured-output validation while maintaining a 'fail-closed' security posture for unverified or failed dimensions.

workflows · high confidence

New Dockerized evaluation harness for context profiles and native probes

A new \docker/context-profiles\ directory introduces a self-contained evaluation environment for testing context profiles and native plugin discovery. It includes a Dockerfile that builds an isolated consumer image (using \node:22-bookworm-slim\) to run \packed-smoke.js\, which validates the full lifecycle of Lean, Full, and Auto context profiles, including native discovery, rollback, and idempotency. The directory also provides a preregistered AI evaluation corpus (\ai-corpus.json\) and an evaluator library (\ai-eval-lib.js\) to grade task success and provider usage across different installation arms (full, manual-lean, auto-lean, ecc-legacy, baseline). Additionally, it adds standalone probe scripts (\native-probe.js\, \native-switch-probe.js\) for exercising real native discovery without creating model threads or copying credentials.

docker/context-profiles · high confidence

New Gateguard evaluation scenarios for agent task effectiveness

Added five new evaluation scenarios under docker/gateguard-effectiveness/scenarios to measure how effectively trapped agent tasks are handled. The scenarios cover: data-schema-orders (parsing day-first dates with semicolons and decimal commas), existing-tests-discount (updating discount logic while maintaining existing tests), importers-date-shape (changing parseDate return type from object to Date), local-callers-normalize (refactoring normalize to return a Map), no-duplicate-slug (creating tag slugs using existing text utilities), and secrets-config-key (moving plaintext secrets to environment variables). Each scenario includes a task definition, workspace starter code, naive/reference implementations, and a grader to score agent performance.

(repo-wide) · high confidence

New Kiro IDE hooks for automated code quality and safety checks

The \.kiro/hooks\ directory now includes a set of pre-configured IDE hooks that automatically trigger agent actions based on file changes and tool usage. These hooks provide immediate feedback for TypeScript/JavaScript formatting, type checking, and console.log cleanup; Python linting; Rust compilation checks; and security reviews for sensitive directories. Additional hooks enforce TDD practices, prevent unnecessary documentation, review git push operations, and generate session summaries or lessons-learned patterns upon agent completion. A manual quality gate hook is also available to run build, lint, and test scripts on demand.

.kiro/hooks · high confidence

New OpenCode tools for testing, linting, formatting, and security auditing

The .opencode/tools directory now includes a suite of custom OpenCode tools that extend the IDE's capabilities. Users can now run tests with automatic package manager and framework detection (run-tests), check test coverage against thresholds (check-coverage), and view changed files as a navigable tree with change indicators (changed-files). Code quality is supported via lint detection and command generation for linters like Biome, ESLint, and Ruff (lint-check), as well as formatter detection and command generation for Biome, Prettier, Black, and others (format-code). Dependency management is enhanced with a dependency analyzer that identifies outdated or vulnerable packages (dependency-analyzer), and security is improved with a security audit tool that scans for dependency vulnerabilities, hardcoded secrets, and code anti-patterns (security-audit). Additionally, a git summary tool provides branch status, recent commits, and diff stats (git-summary). These tools are exported via a central index and are designed to be robust, with lazy loading to prevent session crashes if plugin dependencies are missing.

.opencode/tools · high confidence

New agent skills for debugging, selective installation, and API design

Added three new skills to the agent toolkit: \agent-introspection-debugging\ provides a structured four-phase workflow (capture, diagnose, recover, report) for systematically debugging AI agent failures; \agent-sort\ enables evidence-backed selective installation by classifying ECC components into daily-use and library buckets based on repository context; and \api-design\ offers comprehensive REST API design patterns including resource naming, status codes, pagination, and error handling conventions.

.agents/skills · high confidence

New audit retention purge scenario with hidden 400-day intent

A new evaluation scenario named 'audit-retention-purge' has been added to the gateguard-effectiveness suite. This scenario tests an agent's ability to correctly implement a \purgeAudit\ function that marks records older than a 400-day window with a \deletedAt\ timestamp while leaving them in the array. The scenario is designed to be ambiguous: the visible configuration (\config/retention.json\) only specifies a 30-day session window, hiding the actual 400-day audit retention requirement. The grader (\grader.cjs\) validates the implementation against this hidden intent by checking that records aged 401 and 800 days are marked as deleted, while those aged 10, 100, and 399 days remain untouched.

docker/gateguard-effectiveness/scenarios-intent/audit-retention-purge · high confidence

New development harness for measuring GateGuard question effectiveness

Added a new \docker/gateguard-effectiveness\ directory containing a development-only harness to evaluate how well GateGuard's questions obtain decisive facts. The harness includes two runners: \run.js\ for measuring the gate's turn and token cost in one-shot autonomous sessions, and \run-intent.js\ for measuring effectiveness in multi-turn sessions where the deciding fact exists only in a simulated user's hidden intent. The system supports multiple experimental arms (off, gate, placebo, minus-target, main) and classifies outcomes into descriptive hole categories (e.g., silence-hole, coverage-hole) rather than asserting coverage. It includes grader validation to ensure graders check content, not just form, and generates an evidence manifest with SHA-256 checksums to verify the integrity of run artifacts.

docker/gateguard-effectiveness · high confidence

New opt-in AURA trust-check adapter for agent reputation

The \integrations/aura\ module introduces a zero-dependency, read-only adapter that queries the AURA Open Protocol to assess the counterparty reputation of an agent before sensitive actions like settlement or delegation. It provides an \aura\_verdict\ function for inspection and a \before\_settle\ gate that raises \AuraUntrusted\ by default for agents with no history (\new\) or poor track records (\high\_risk\), while allowing \trusted\ and \caution\ agents to proceed. The adapter is off by default, requires no authentication, and includes a threat model and offline tests to ensure safe integration.

integrations · high confidence

New scripts for agent conversion, auto-update, and install catalog management

Added several new CLI scripts to the \scripts/\ directory to support the expanded harness ecosystem. \scripts/agent-convert.js\ provides a command-line tool to convert ECC's canonical agent definitions from the Claude format to the Pi format, including validation (\--check\) and dry-run capabilities. \scripts/auto-update.js\ introduces a guided auto-update mechanism that pulls the latest repository changes and reinstalls managed targets while preserving user consent and configuration state, with strict validation to prevent execution from untrusted repository roots. \scripts/catalog.js\ exposes a CLI for discovering and inspecting install components, profiles, and their metadata. Additionally, \scripts/build-opencode.js\ was added to handle the TypeScript compilation of the OpenCode plugin assets.

scripts · high confidence

New skills for accessibility auditing, agent evaluation, and self-debugging

Added new skill definitions for accessibility (WCAG 2.2 compliance across Web, iOS, and Android), agent architecture auditing, head-to-head coding agent evaluation, agent harness construction, and structured agent introspection debugging. These skills provide structured workflows and best practices for ensuring inclusive design, diagnosing agent failures, and comparing agent performance.

skills · high confidence

New unified memory conformance example validates CLI and MCP evidence integrity

A new example in examples/unified-memory validates that the ECC CLI and local stdio MCP server handle memory records consistently and securely. The conformance runner (conformance.cjs) creates disposable synthetic vaults to verify that scoped queries return identical ordered records, scores, and provenance across CLI and MCP handoffs, while enforcing strict access controls such as rejecting client identity overrides, target-filter spoofing, and unauthorized trust promotion. It also introduces a new evidence verification module (evidence.cjs) that validates the integrity of memory bodies against a host-owned in-memory catalog, ensuring source-content matching and rejecting tampered, missing, or foreign-context evidence. The example includes dedicated boundary tests (evidence.test.cjs) for schema validation, digest integrity, and control character rejection, all running without external services or network dependencies.

examples/unified-memory · high confidence

New workflow for ECC Pro security roadmap planning

A new workflow file has been added to automate the quarterly planning process for ECC Pro and AgentShield security. This workflow guides the agent through a three-phase process: surveying current capabilities and triaging open issues, researching external threats and competitor gaps, and synthesizing a prioritized, revenue-biased security roadmap. It includes specific schemas and guardrails to ensure the output focuses on monetizable features and addresses critical trust issues like false positives.

.claude/workflows · high confidence

Behavioural changes

Audit retention logic now distinguishes between hard delete, soft delete, and no retention

The audit retention behavior for the 'audit-retention-purge' scenario is now implemented across three distinct configurations. The 'naive' implementation performs a hard delete of records older than the configured session window. The 'reference' implementation uses a soft-delete approach, marking records with a 'deletedAt' timestamp after 400 days while keeping them readable for auditors. The 'workspace' implementation currently retains all records indefinitely without any purge or soft-delete logic.

docker/gateguard-effectiveness/scenarios-intent/audit-retention-purge/naive, docker/gateguard-effectiveness/scenarios-intent/audit-retention-purge/reference, docker/gateguard-effectiveness/scenarios-intent/audit-retention-purge/workspace · high confidence

Decoy script mutates input rows before CSV export

The decoy script in the CSV export delimiter scenario now overwrites the values of all input rows with 'x' before generating the CSV output. This ensures that the original data is lost and replaced with dummy values, simulating a scenario where data integrity is compromised during export.

docker/gateguard-effectiveness/scenarios-intent/csv-export-delimiter/decoy-mutates · high confidence

Updated brand assets and documentation imagery for ECC v2.0.0

The repository's visual identity has been refreshed with new SVG assets: a new \ecc-icon.svg\ replaces the previous brand mark, and a redesigned \hero.svg\ banner now displays the ECC v2.0.0 branding, updated taglines, and a catalog of supported agents and skills. Additionally, \star-history-data.tsv\ has been added to provide the underlying data for GitHub star history visualizations in the documentation.

assets · high confidence

Fixes

107 commits (64 fixes) fixing hooks

A fix in hooks — 107 commits (64 fixs), 9 files: “refactor(hooks): consolidate PostToolUse hooks into sync/async dispatchers (\#2494)”; “fix: resolve blocker PR validation regressions”; “fix(skill-evolution): wire Skill PostToolUse tracker so skill-health shows real runs (\#2490)”; and 104 more (refs \#2494, \#2490, \#491, \#485, \#482, …).

hooks · medium confidence · unverified

38 commits (20 fixes) fixing pi

A fix in pi — 38 commits (20 fixs), 209 files: “chore(release): 2.2.3, ECC-only naming in pi/core and a longer npm publish wait (\#3329)”; “fix: close installed workflow and review publication gaps before integration”; “feat(skills): add context and decisions”; and 35 more (refs \#3329, \#3264).

pi · medium confidence · unverified

Fix quoting bypasses and preserve Git configuration in shell hooks

The shell hook scanning logic has been refactored to include a dedicated finite literal scanner (shell-scan.js) that correctly handles heredoc delimiters, quote removal, and command substitutions, closing previous quoting bypasses. Additionally, the hooks now preserve Git configuration across different shell command scopes and strictly detect literal hook bypasses, ensuring that security checks remain effective even when complex shell syntax is involved.

scripts/hooks · high confidence

Fixes data directory resolution logic for continuous learning

The continuous learning scripts now correctly resolve the data directory by validating that environment variables (CLV2\_HOMUNCULUS\_DIR, XDG\_DATA\_HOME) and HOME are absolute paths, falling back to a default location only when necessary, which prevents errors caused by relative path configurations.

skills/continuous-learning-v2/scripts · high confidence

Test coverage

Add CI validation tests for agent instruction safety, command registry, and skill surfaces; Add decoy CSV export modules for testing scenarios; Added evaluation scenarios for retry idempotency, timezone rollup, and username collation; Added gateguard effectiveness scenarios for intent detection; Added grader and scenario for CSV export delimiter validation.

Dependencies

Initial dependency manifests for ECC 2.2.2 across Node, Rust, and Python

This change introduces the initial dependency lockfiles and manifests for the ECC 2.2.2 release, establishing the baseline for the project's three main runtime environments. For the Node.js harness, it adds \package.json\ and \package-lock.json\ (root and \.opencode/\) pinning \typescript\ to 5.9.3/6.0.3, \@opencode-ai/plugin\ to 1.4.3/1.18.25, and \ajv\ to 8.20.0, alongside dev tools like \eslint\ 10.9.1 and \c8\ 11.0.0. For the Rust TUI control plane (\ecc2\), it adds \Cargo.toml\ and \Cargo.lock\, specifying \ratatui\ 0.30, \crossterm\ 0.29, \rusqlite\ 0.40, and \git2\ 0.21. For the Python LLM abstraction layer, it adds \pyproject.toml\ requiring \anthropic\>=0.120.2\ and \openai\>=2.34.0\, along with test dependencies like \pytest\>=9.1.1\ and \mypy\>=2.3.0\. It also includes initial \requirements.txt\ files for media skills (e.g., \taste-application\) and a test fixture \package.json\.

(dependencies) · high confidence

Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

How this codebase got here

Score

  • CAI 52 → 52 (-0.6)
  • Rubric changed (rubric-2026.10.5 → rubric-2026.11.3) — scores are not directly comparable.

Lenses

  • Code Health 42 → 42 (+0.1)
  • Architecture 92 → 92 (+0.0)
  • Maturity 85 → 95 (+10.2)
  • Readiness 53 → 48 (-5.1)
  • Security 68 → 66 (-1.7)
  • Event Sourcing 100 → 100 (+0.0)
  • Performance 60 → 60 (+0.0)

Resolved (734)

  • Documentation: written for insiders (.trae/README.md)
  • Documentation: written for insiders (ecc2/README.md)
  • Duplicated block (10 lines × 2) (ecc2/src/main.rs)
  • Duplicated block (10 lines × 2) (ecc2/src/session/manager.rs)
  • Duplicated block (10 lines × 2) (ecc2/src/session/store.rs)
  • Duplicated block (10 lines × 2) (pi/core/skills/agent-self-evaluation/scripts/evaluate.py)
  • Duplicated block (10 lines × 2) (pi/core/skills/agent-self-evaluation/scripts/evaluate.py)
  • Duplicated block (10 lines × 2) (skills/taste-application/scripts/taste/frames.py)
  • Duplicated block (10 lines × 2) (skills/taste-application/scripts/taste/grade.py)
  • Duplicated block (11 lines × 2) (ecc2/src/worktree/mod.rs)
  • Duplicated block (11 lines × 4) (ecc2/src/worktree/mod.rs)
  • Duplicated block (11–12 lines × 2) (ecc2/src/main.rs)
  • Duplicated block (12 lines × 2) (ecc2/src/main.rs)
  • Duplicated block (12 lines × 2) (pi/core/skills/agent-self-evaluation/scripts/evaluate.py)
  • Duplicated block (12 lines × 2) (skills/taste-application/scripts/taste/grade.py)
  • Duplicated block (12 lines × 3) (skills/taste-application/scripts/falapi.py)
  • Duplicated block (13 lines × 2) (skills/taste-application/scripts/mint.py)
  • Duplicated block (14 lines × 2) (skills/taste-application/scripts/taste/frames.py)
  • Duplicated block (14 lines × 2) (skills/taste-application/scripts/taste/grade.py)
  • Duplicated block (15 lines × 2) (ecc2/src/session/manager.rs)
  • …and 714 more

New (179)

  • Duplicated block (10 lines × 2) (ecc2/src/main.rs)
  • Duplicated block (10 lines × 2) (ecc2/src/session/manager.rs)
  • Duplicated block (10 lines × 2) (ecc2/src/session/store.rs)
  • Duplicated block (11 lines × 2) (ecc2/src/session/store.rs)
  • Duplicated block (11 lines × 2) (ecc2/src/worktree/mod.rs)
  • Duplicated block (11 lines × 4) (ecc2/src/worktree/mod.rs)
  • Duplicated block (11–12 lines × 2) (ecc2/src/main.rs)
  • Duplicated block (12 lines × 2) (ecc2/src/main.rs)
  • Duplicated block (13–14 lines × 4) (ecc2/src/session/store.rs)
  • Duplicated block (13–15 lines × 2) (ecc2/src/tui/dashboard.rs)
  • Duplicated block (14 lines × 2) (ecc2/src/tui/dashboard.rs)
  • Duplicated block (15 lines × 2) (ecc2/src/session/manager.rs)
  • Duplicated block (16 lines × 3) (.agents/skills/docs-governance/scripts/audit-docs.py)
  • Duplicated block (16–18 lines × 2) (ecc2/src/session/manager.rs)
  • Duplicated block (49 lines × 3) (.agents/skills/docs-governance/scripts/module_audit.py)
  • Duplicated block (5 lines × 2) (ecc2/src/session/store.rs)
  • Duplicated block (7 lines × 2) (ecc2/src/worktree/mod.rs)
  • Duplicated block (8 lines × 2) (ecc2/src/session/manager.rs)
  • Duplicated block (8–10 lines × 2) (ecc2/src/worktree/mod.rs)
  • Duplicated block (9–13 lines × 2) (ecc2/src/tui/dashboard.rs)
  • …and 159 more

Changes since last survey

  • 251 commits — 142 feature/other, 109 fixes

By area

  • (repo) — 87 commits
  • scripts/lib — 21 commits
  • ecc2/src — 19 commits
  • docs/ja-JP — 13 commits
  • manifests/context-packs — 13 commits
  • scripts/hooks — 11 commits
  • examples/evaluator-rag-prototype — 9 commits
  • tests/hooks — 8 commits
  • tests/scripts — 8 commits
  • (root) — 6 commits
  • tests/lib — 6 commits
  • .github/workflows — 4 commits
  • skills/skill-comply — 4 commits
  • .codex-plugin/README.md — 3 commits
  • docs/QWEN-GUIDE.md — 3 commits
  • pi/core — 3 commits
  • skills/continuous-learning-v2 — 3 commits
  • skills/videodb — 3 commits
  • scripts/codex — 2 commits
  • scripts/gemini-adapt-agents.js — 2 commits

Notable commits

  • fix: Merge pull request #3583 from Tivonsico/fix/skill-run-timestamp-type-20261014
  • fix: Merge pull request #3594 from ketpatil77/fix/managed-hook-config-bom
  • fix: Merge pull request #3650 from haelyra/fix/pr-open-cap-grandfather
  • fix: Merge reviewed OpenCode JSONC fix from #3590
  • fix: chore: sync fractional timestamp fix with main
  • fix: fix(ci): verify pinned Windows ripgrep before running test gates
  • fix: fix(ci): verify pinned Windows ripgrep without Chocolatey
  • fix: fix(ck): never let forget remove the shared contexts root
  • fix: fix(ck): reject only parent-traversal segments in the forget guard
  • fix: fix(codex): bind cache rerun guidance to active plugin
  • fix: fix(codex): recommend native cache recovery in checker
  • fix: fix(config-protection): check MultiEdit edits[].file_path targets
  • fix: fix(context): bind curated triggers to combined reviewed registry
  • fix: fix(context): bind final reviewed guidance registry
  • fix: fix(context): rebind registry after verified review repairs
  • fix: fix(context): validate budgets and remind before native compaction
  • fix: fix(continuous-learning-v2): redact cloud keys, DSNs and PEM blocks in observations
  • fix: fix(continuous-learning-v2): redact numeric DSN passwords cut at the truncation boundary
  • fix: fix(continuous-learning-v2): redact tokens after JSON escapes and truncated DSNs
  • fix: fix(deps): integrate rustls 0.23.46 security update (#3807)
  • …and 231 more

Architecture

  • Unchanged — 0 containers · 1 contexts · 0 edges

Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

Survey your own repository

affaan-m/ECC was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.

About this page

  • The score is its most recent published measurement, taken on 13 November 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
  • Measured at commit d2ef2680594f9a17526a6e1a2e18b3d242c5d17b — the exact code this score is about.
  • Scored under rubric-2026.11.3 — the same rubric and the same method as every other entry in this index.
  • Measured by Canine Code Analyzer (cca.canine.dev) using cca measure-2026.11.61+preprod-f3449e6eb431.