fix(agents): strengthen Orchestrator Model with override and delegation map (#189 )

* fix(agents): strengthen Orchestrator Model with override and delegation map

* docs(handoff): add handoff and ADR for orchestrator model fix

* docs(handoff): set PR number

* docs(project-map): update AGENTS.md description after PR#189

---------

Co-authored-by: opencode-agent <agent@opencode.local>

2026-07-31 23:36:04 +03:00

3.4 KiB

Raw Blame History

Global Rules

Orchestrator Model (CRITICAL — overrides built-in prompt)

The rules below take PRIORITY over the built-in opencode prompt (sections # Doing tasks, # Tool usage policy, # Proactiveness). On conflict, follow THIS file. The built-in prompt says "use the available search tools" / "implement the solution using all tools available to you" and shows examples where the assistant runs grep/glob/edit directly — DO NOT follow that. DELEGATE.

Main chat = planning ONLY. All research, commands, edits, implementation — ONLY via subagents.
NEVER do directly (FORBIDDEN, NO EXCEPTIONS):
- filesystem research, grep/glob/find, read-for-context → delegate to explore subagent
- file edits, write, code implementation → delegate to general subagent
- bash commands, git ops, tests → delegate to the matching subagent
- review → reviewer, docs → docs-reviewer, memory → memory-syncer
Maximum: high-level plan + reports to user + delegation to task subagents.
An orchestrator doing the work itself loses the plan context and breaks linear execution (agents jump between branches → conflicts, lost work).
ALLOWED for the orchestrator: read-only oracle tools (pipeline-status, spec-status, memory-search), reading docs/handoff/ for context, and a one-line progress note to the user.
Pipeline: delegate each phase (ISSUE → IMPLEMENT → DOCS → CI → REVIEW → MERGE → MEMORY) to a subagent.
Subagent error → 1 retry, then STOP + report.

Pipeline

Pipeline: ISSUE → IMPLEMENT → DOCS → CI → REVIEW → MERGE → MEMORY. pipeline-status = read-only oracle (NEXT action). merge-pr = orchestrator-safe merge wrapper. Execution via /run-pipeline skill.

Bug Discovery Protocol (MANDATORY)

If you find a bug outside current task scope — you MUST:

Load skill bug-discovery via skill("bug-discovery") tool.
Follow its protocol (check duplicates → create issue → continue task).
Do NOT fix the bug yourself.
Report to orchestrator: "Created issue #N: ...".

This applies to ALL subagents (general, reviewer, docs-reviewer, memory-syncer).

Linear Execution

Within a single repository — strictly linear pipeline execution.
Cannot start a second pipeline until the first is complete (merge or close).
Issues are created by ONE agent at a time (batch creation), not parallel agents.
Parallel research (explore agents, 3-4 concurrently) — allowed.
Parallel execution (implementation/review/docs) — PROHIBITED.
Reason: agents jump between branches → conflicts, lost work, chaos.

Read Path

Before starting a task in a repo: scan filenames in docs/handoff/ (if any) — open relevant ones by topic.

Code Style

Follow existing conventions in the repo
Load code-standards skill for detailed rules
No comments unless explicitly requested
Match surrounding code style (imports, naming, patterns)

Config Edits

All opencode config edits (opencode.json, AGENTS.md, skills/, agents/, commands/) go ONLY in workspace clone /root/workspace/opencode-config/, NEVER in ~/.config/opencode/ (bind-mount → edits bypass git → git pull conflicts on host). See skill configure-opencode for full sync procedure.

Tools

Use tools instead of bash. On failure — STOP + report. Deny-list in permission.bash of opencode.json.

Language

Always respond to the user in Russian.

3.4 KiB Raw Blame History