Skip to content

v1.15.0 — Orchestrator-Centric Execution, Single-Sourced Prompts, Skills-Only

Latest
Latest

Choose a tag to compare

View all tags
@zhu1090093659 zhu1090093659 released this 26 Jul 11:19
· 1 commit to main since this release
v1.15.0
6fdd48d
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

v1.15.0

A workflow-focused release: the execution model becomes orchestrator-centric with tiered sub-agent dispatch and review, all prompts are single-sourced and concise, and the plugin drops the redundant commands surface to become skills-only.

Highlights

  • Orchestrator-centric execution model — the main agent owns progress and quality. Sub-agent dispatch is now an economic decision, not a default: orchestrator-direct execution (Tier 0) is the norm, a single task-executor coder (Tier 1) covers large/context-heavy batches, and parallel lanes (Tier 2) require disjoint file sets, ≥ L effort per lane, independent verifiability, and ≤ 4 lanes. This directly addresses subscription-quota burn from over-dispatching.
  • Per-lane independent review — a new code-reviewer sub-agent verifies each lane's diff against its acceptance criteria, commits fixes directly to the lane branch (append-only fix: commits), and returns a structured verdict (APPROVED / FIXED / ESCALATE). Review is likewise tiered: machine validation (L1) and the orchestrator's own diff review (L2) by default; the independent reviewer (L3) is reserved for Tier 2 lanes and high-risk changes. Writer-model invariants keep the orchestrator the single writer for all shared state.
  • Single-sourced, concise prompts — SKILL.md restructured from 405 → 249 lines with a canonical-reference table; every cross-reference uses named anchors instead of fragile section numbers; each topic (rules, GitHub schemas, telemetry, tiers, S.U.P.E.R, templates) has exactly one home. Satellite skills slimmed: deep-discuss −61%, review-spd tightened. Agent input/output contracts preserved byte-identical.
  • Skills-only surface — the commands/ directory is removed atomically across all 5 surfaces (plugin manifests, opencode plugin, CLAUDE.md, both READMEs). Workflows are invoked purely through skills.
  • scripts/install-agents.sh — new installer syncs the bundled skills to ${AGENTS_HOME:-~/.agents}/skills, wired into install-all.sh.
  • Validation gate — scripts/validate.sh checks reference existence, manifest↔filesystem parity, 4-site version parity, JSON validity, ESM syntax, py_compile, and exporter smoke.

Upgrade notes

  • The /spec-dev and /dp slash commands are gone — invoke the spec-driven-develop and deep-discuss skills instead.
  • No breaking changes to generated document formats; export-progress.py-parsed progress patterns are unchanged (now formally frozen).

Closed issues: #14–#38 across 5 batch PRs (#40–#44).

Assets 2
Loading