Epic: Conversational orchestrator: hold context, surface gaps, propose improvements like a teammate

Open
  • Is something wrong?
#44 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 6 days

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
12/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Active
Tech stack
javascript
Domain
ai, devtools

Research direction

This is a tracking epic, not a starting point: it spans hooks in bundle/hooks.json, the memory modules in lib/memory-manager.js, and the planner, verifier and reviewer prompts. Pick a child task with a narrower scope, such as #49 (report format) or #50 (config flags in lib/orchestra-config.js). Done means that child's acceptance criteria pass and the existing suites stay green with all flags off.

Written by the indexing model from the issue text.

Description

enhancement epic

Motivation

Today pcore-orchestra runs a strong plan → implement → verify → review loop per prompt, then forgets. Users want an orchestrator that behaves like a teammate: it remembers earlier decisions, notices gaps in the project, asks when a request is unclear, and proposes the next improvements — while staying opt-out-able and within token budgets.

Tasks

  • #45 Persistent conversation/context store across sessions
  • #46 Self-audit mode: on-demand or scheduled project audit that reports gaps
  • #47 Proactive suggestions: reviewer emits "Next improvements" after each run
  • #48 Clarifying-question step for ambiguous prompts
  • #49 Audit report output format and save location (good first issue)
  • #50 Config and opt-out flags for proactive behaviour (good first issue)
  • #39 Command-triggered research: orchestrator researches and updates its own knowledge before working (good first issue)

Related (not part of the epic's scope but interacts): #40 token reduction, #42 monorepo support.

Design (ties into existing code)

  • Hooks: bundle/hooks.json events sessionStart, beforeSubmitPrompt, subagentStop, stop → orchestra-session-start.js, orchestra-prompt.js, orchestra-subagent-stop.js. Context recall happens at session start / prompt; suggestions and decisions are captured at subagentStop/stop.
  • Memory / recall: lib/memory-manager.js, memory-index.js, memory-fts.js, semantic-memory.js (FTS5 on Node ≥ 22.5, JSON fallback) become the storage for the context store, research cache and suggestion de-duplication.
  • Planner stage (orchestra-planner.md): clarifying questions (#48) and research (#39) run before PLAN_READY.
  • Verifier stage: checks the generated project baseline checklist (#39) and records gaps.
  • Reviewer stage (orchestra-reviewer.md): emits "Next improvements" (#47) alongside ORCHESTRA_COMPLETE / FIX_REQUIRED.
  • Self-audit (#46) reuses evidence collection (lib/evidence-runner.js) and writes the report format from #49.
  • Config: project.yaml → lib/orchestra-config.js; flags from #50 gate every feature above.

Per-host feasibility

Host Hooks available How the features run
Cursor ~/.cursor/hooks.json (4 events) Full: recall at prompt/session start, capture at subagentStop/stop
Claude Code settings.json hooks (SubagentStop can block+reason) Full; block+reason can deliver follow-up questions/suggestions
Qoder settings.json hooks (PascalCase events) Full (shared hook scripts via lib/io.js shim)
OpenCode none; conductor + orchestra-ambient.js plugin Partial: conductor instructions + ambient plugin context; store read/written via lib CLI
Codex none (hooks gated by features.hooks; installer leaves unwired) Instruction-only via AGENTS.md/agents; no automatic capture — document limitations

Out of scope

Cloud sync of conversation data; autonomous execution of suggestions without user approval.

Acceptance criteria

  • All tasks above are closed (or explicitly descoped with a note here).
  • With every flag off (#50) behaviour is identical to v3.8.0 (existing suites pass).
  • End-to-end demo: a second session recalls a decision from the first, asks one clarifying question on an ambiguous prompt, ends with a "Next improvements" list, and /orchestra audit writes a report in the documented format.
  • Per-host behaviour documented (including the Codex/OpenCode limitations); token overhead of the added context measured against the budgets (see #40, #41).
  • No secrets stored in the context store (security scan test).

Raised from the v3.8.0 audit (commit 22d2146).

Dominant language
JavaScript
Stars
0
Forks
0
No merged PRs in 30d

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from P-Core-System/pcore-orchestra

All issues in P-Core-System/pcore-orchestra

Similar issues

More JavaScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.