Skip to content

roadmap: detailed feature specs for v1.1 and v1.2#76

Merged
emal-avala merged 4 commits intomainfrom
roadmap/detailed-feature-additions
Apr 6, 2026
Merged

roadmap: detailed feature specs for v1.1 and v1.2#76
emal-avala merged 4 commits intomainfrom
roadmap/detailed-feature-additions

Conversation

@emal-avala
Copy link
Copy Markdown
Member

Summary

Adds comprehensive engineering specifications for 15 new features to ROADMAP.md, organized into two release targets (v1.1 and v1.2).

v1.1 — Critical & High Priority

  • Process-level sandboxing (7.4) — OS-level isolation via seatbelt/bubblewrap/Windows integrity
  • Behavioral eval framework (7.5) — Tests agent behavior with live LLMs, best-of-N retry, auto-promotion
  • Model routing & fallback (4.3) — Availability state machine with automatic degradation
  • Epic workflows (7.12) — Stateful plan-based implementation with worktree isolation
  • Provider prompt caching (7.13) — Anthropic cache_control, up to 90% cost reduction
  • Advanced compression (7.6) — File-level tracking with content hashing and secret masking
  • JSON streaming output (7.7) — JSONL format for CI/CD with structured exit codes
  • Hierarchical context (7.17) — Monorepo support via nested context file merging
  • Shell passthrough (7.18) — ! prefix for direct commands with context injection

v1.2 — Medium Priority

  • Remote agent protocol (7.8) — JSON-RPC 2.0 with Agent Card discovery
  • DevTools server (7.9) — WebSocket debugging with bounded event buffers
  • GitHub Action (7.10) — PR review and issue triage automation
  • OAuth + free tier (7.11) — Browser-based auth with OS keychain
  • Local LLM discovery (7.14) — Auto-detect Ollama/LM Studio
  • Conversation branching (7.15) — Fork, rewind, merge conversation states
  • Security scanning (7.16) — Dependency audit + code pattern detection

Each feature includes architecture diagrams, config schemas, implementation checklists, and test requirements.

Test plan

  • Renders correctly in GitHub markdown
  • No competitor names mentioned
  • Implementation tasks reference correct crate paths
  • Team review of priority ordering

🤖 Generated with Claude Code

@chatgpt-codex-connector
Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

@emal-avala emal-avala force-pushed the roadmap/detailed-feature-additions branch from 2777e2f to c6dd089 Compare April 6, 2026 09:00
Add comprehensive engineering specifications for 15 new features spanning
security, testing, developer experience, and ecosystem expansion:

v1.1 (Critical/High):
- Process-level sandboxing (macOS seatbelt, Linux bubblewrap, Windows integrity)
- Behavioral evaluation framework with best-of-N retry and auto-promotion
- Model routing with availability state machine and fallback chains
- Stateful epic workflows with worktree-isolated implementation
- Provider prompt caching (Anthropic cache_control, cost tracking)
- Advanced history compression with file-level tracking and secret masking
- Non-interactive JSON streaming output for CI/CD pipelines
- Hierarchical context resolution for monorepos
- Shell passthrough mode (! prefix)

v1.2 (Medium):
- Remote agent protocol (JSON-RPC 2.0, Agent Card discovery)
- DevTools debugging server (WebSocket, bounded event buffers)
- GitHub Action for automated PR review and issue triage
- OAuth authentication with OS keychain and free tier
- Local LLM auto-discovery (Ollama, LM Studio, llama.cpp)
- Conversation branching (fork, rewind, merge sessions)
- Proactive security scanning (dependency audit, code pattern detection)

Each feature includes architecture diagrams, configuration schemas,
implementation task checklists, and testing requirements.
…esktop client specs

- Replace all provider-specific model names with generic placeholders
- Replace provider-specific API key names with generic LLM_API_KEY
- Rename "Provider Prompt Caching" to be provider-agnostic
- Fix 7.18: clarify shell passthrough already partially implemented (! prefix
  exists in repl.rs:603), spec now covers context injection and real-time
  streaming upgrades only
- Fix 7.21: change desktop client from Tauri to Flutter per team preference,
  add JSON-RPC over WebSocket IPC per prior architectural decision
- Add missing 4.3 Model Routing section (lost during file edit race)
- Add gVisor as future Linux sandbox strategy in 7.4
- Add cross-references between routing (4.3) and caching (7.13) for
  cache invalidation on model switch
- Remove free tier from 7.11 OAuth (requires server infra, out of scope)
- Move 7.17 and 7.18 from v1.1 to v1.2 (medium priority != v1.1 target)
- Fix routing strategy ordering: OverrideStrategy now first (user /model
  override must always win, not be silently ignored by FallbackStrategy)
- Fix sandbox wiring point: wrap at Command::new sites (bash.rs, agent.rs,
  powershell.rs) not executor.rs (which calls trait methods, not subprocesses)
- Add SandboxedCommand wrapper concept
- Expand secret masking scope to all persistence points: session.rs,
  output_store.rs, /share export (not just compression path)
@emal-avala emal-avala force-pushed the roadmap/detailed-feature-additions branch from c6dd089 to 0df4d8d Compare April 6, 2026 09:01
@emal-avala emal-avala merged commit 83e89db into main Apr 6, 2026
13 of 14 checks passed
@emal-avala emal-avala deleted the roadmap/detailed-feature-additions branch April 6, 2026 09:02
emal-avala added a commit that referenced this pull request Apr 6, 2026
PR #78 squash merge accidentally removed the 15 feature specs added by
PR #76. This restores them and applies accuracy corrections from an
audit against the actual codebase:

- 7.13 Provider Prompt Caching: marked Done — system prompt caching,
  message breakpoints, cache tracking, cost display, tool caching,
  and config toggle all implemented
- 7.14 Local LLM Auto-Discovery: marked Partially Done — Ollama
  detection already exists in setup.rs, only LM Studio/llama.cpp
  remaining
- 7.15 Conversation Branching: marked Partially Done — /fork command
  and /resume exist, advanced branching (named branches, checkout,
  merge) still needed
- Updated Contributing section to reflect completed items
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant