docs(goals): adopt advisor + plan-big/execute-small patterns in agent personas - #28
Merged
Merged
Conversation
… personas
Reviewed the goal org against Anthropic's advisor tool and the plan-big/execute-small
managed-agents cookbook. The org already is the coordinator pattern (fable/opus plan,
sonnet execute, per-worktree context isolation, distilled reports up) and is ahead on
durability. Imports the rest as persona-prompt edits:
* Advisor consult (advisor tool): an impl agent (sonnet) that hits a strategic fork
consults its live spec agent (opus/fable) over the existing message bus
(send_message/read_messages) and continues on sonnet — opus-grade decisions at
sonnet rates, instead of guessing or a full nack->re-spec round-trip. Spec agent
stamps advisor=<paneId> on each payload; the same channel runs spec->orchestrator,
and the orchestrator answers as the top-tier advisor.
* Don't decompose when it doesn't pay (cookbook #9): small/atomic goals built inline.
* Right-size briefs / delegation floor cost (cookbook #7): batch trivia, don't spawn a
worktree agent per two-line edit.
* Verify the premise, not just the artifact (cookbook #8): spec agent sanity-checks its
own decomposition before fan-out.
* Synchronization barrier (cookbook #4): wait for the whole wave before verifying/done.
SKILL/SPEC/IMPL.md + a rationale section in goals-system-plan.md. Prompt-only; no code.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Checked the goal orchestrator against two Anthropic patterns: the advisor tool and the plan-big/execute-small managed-agents cookbook.
Verdict: the org already is the coordinator pattern — fable/opus plans, sonnet executes, context isolated per worktree pane, distilled reports up — and on durability (SQLite queue with leases/backoff/dead-letter, session-daemon resume, account rotation) it's ahead of both docs. So this isn't a rewrite; it imports the five ideas we were missing, all as persona-prompt edits.
What
send_message/read_messages) and continues on sonnet — opus-grade calls at sonnet rates, instead of guessing or a coarse nack→re-spec round-trip. Spec agent stampsadvisor=<paneId>on each payload; the same channel runs spec→orchestrator, and the orchestrator answers as the top-tier advisor.claude-boot floor is far higher than a web-fetch worker.done.Grounded in real primitives:
send_message {to,from,body}/read_messages {paneId,after}(verified againstmcp-gate/gate.mjs),$HYPERPANES_PANE_IDper pane.Files
resources/claude/goal-orchestrator/{SKILL,SPEC,IMPL}.md+ a rationale section indocs/goals-system-plan.md. Prompt/docs only — no code, no behavior change to the runtime.🤖 Generated with Claude Code