diff --git a/conductor/tracks/superpowers_review_20260619/report.md b/conductor/tracks/superpowers_review_20260619/report.md index 6d95f832..a773cb9a 100644 --- a/conductor/tracks/superpowers_review_20260619/report.md +++ b/conductor/tracks/superpowers_review_20260619/report.md @@ -935,7 +935,140 @@ The project follows the skill's structure but diverges on three points: descript ## 15. MMA Skills Cluster - +## 15. MMA Skills Cluster + +The project's 5 MMA skills are the project's own multi-model architecture for orchestrating AI agent work. They are NOT superpowers-plugin skills; they are project-specific skills in `.agents/skills/`. This cluster section evaluates each one against the corresponding superpowers skill (where applicable) and against the project's actual practice. + +### 15.1 mma-orchestrator (parent skill, Tier 1 + Tier 2 dispatch protocol) + +**What the skill prescribes.** The `mma-orchestrator` skill (135 lines) enforces the 4-Tier Hierarchical MMA via Token Firewalling + sub-agent task delegation. Tier 1 (Orchestrator) for setup/init; Tier 2 (Tech Lead) for execution. Token firewalling: delegate token-heavy or stateless tasks to Tier 3 / Tier 4 via `run_shell_command` (now `scripts/mma_exec.py` or OpenCode Task tool). Critical Prerequisite: don't call `gemini` directly; use `uv run python scripts/mma_exec.py --role `. Includes the Surgical Spec Protocol (AUDIT BEFORE SPECIFYING, GAPS NOT FEATURES, WORKER-READY TASKS, ROOT CAUSE ANALYSIS, REFERENCE DOCS, MAP DEPENDENCIES), Tier 3 dispatch protocol (pre-delegation checkpoint, code-style enforcement, surgical prompts, repeated-failure escalation), Tier 4 QA dispatch (DO NOT analyze raw logs in your own context), Persistent Tier 2 memory (no Context Amnesia for Tier 2), AST Skeleton & Outline Views for context minimization, Cross-Skill Activation patterns. + +**Mapping to project practice.** The skill's 7-section protocol is the project's actual operating procedure for track creation and execution (per `conductor/workflow.md` §"Conductor Token Firewalling" + §"Tier 1 Track Initialization Rules"). The skill references `docs/guide_*.md` files which exist; the skill mentions `mma_exec.py` which is deprecated in favor of OpenCode Task tool (per `conductor/workflow.md` §"Conductor Token Firewalling" §0 update 2026-06-27). + +**Gaps.** The skill mentions `gemini` directly (per the legacy Gemini CLI era); the project's current OpenCode-based workflow uses OpenCode Task tool instead. The skill's anti-patterns section ("DO NOT SKIP A TEST IN PYTEST JUSTS BECAUSE ITS BROKEN...") has typos ("JUSTS", "TRIVAL", "PSUEDO") that suggest hurried authoring. + +**Verdict.** + +| Field | Value | +|---|---| +| **Primary** | `PARITY` | +| **Integration tag** | `INTEGRATED` | +| **Section size** | cluster | +| **Cross-refs** | n/a (project-specific) | + +**Rationale.** The skill is a faithful re-encoding of the project's MMA pattern; the protocol is followed in practice. Two soft gaps: legacy `gemini` references; typos in anti-patterns section. + +**Recommended change.** Update `gemini` references to OpenCode Task tool; fix typos in anti-patterns section. + +### 15.2 mma-tier1-orchestrator (Tier 1) + +**What the skill prescribes.** The `mma-tier1-orchestrator` skill (49 lines) is the "Focused on product alignment, high-level planning, and track initialization" role. Primary Context Documents: all immediate files in `./conductor`; all docs in `./docs`; AST Skeleton summaries of `./src`, `./simulation`, `./tests`, `./scripts`. Architecture Fallback: `docs/guide_architecture.md`, `guide_tools.md`, `guide_mma.md`, `guide_simulations.md`, `guide_meta_boundary.md`. Responsibilities: maintain alignment with product guidelines + definition; define track boundaries + initialize new tracks; set up project environment; delegate track execution to Tier 2. Surgical Spec Protocol: 6 steps (Audit, Spec gaps, Worker-ready tasks, Root cause analysis, Reference docs, Map dependencies). Limitations: Do not execute tracks or implement features; Do not write code or perform low-level bug fixing; Keep context focused on product definitions + strategy. + +**Mapping to project practice.** The Tier 1 Orchestrator role is the agent currently doing this track (per the developer policy "STRICT SYSTEM DIRECTIVE: You are a Tier 1 Orchestrator"). The skill's Surgical Spec Protocol is enforced via `conductor/workflow.md` §"Tier 1 Track Initialization Rules". The Architecture Fallback list (5 deep-dive guides) is followed. + +**Gaps.** The skill's description is "Focused on product alignment, high-level planning, and track initialization" — persona-flavored, not trigger-focused (per Section 14's writing-skills gap). The "Limitations" section is brief; could include more explicit "do not" examples. + +**Verdict.** + +| Field | Value | +|---|---| +| **Primary** | `PARITY` | +| **Integration tag** | `INTEGRATED` | +| **Section size** | cluster | +| **Cross-refs** | n/a (project-specific) | + +**Rationale.** The skill is the actual operating procedure for Tier 1; followed in practice. The description and limitations could be tighter. + +**Recommended change.** Rewrite description to be trigger-focused ("Use when initializing a new track or planning high-level product alignment"); expand Limitations with specific anti-examples. + +### 15.3 mma-tier2-tech-lead (Tier 2) + +**What the skill prescribes.** The `mma-tier2-tech-lead` skill (53 lines) is the "Focused on track execution, architectural design, and implementation oversight" role. Architecture: YOU MUST READ THE FOLLOWING BEFORE IMPLEMENTING TRACKS — all immediate files in `./conductor`; AST Skeleton summaries of `./src`, `./simulation`, `./tests`, `./scripts`; the 5 deep-dive guides. Responsibilities: manage execution of implementation tracks; ensure alignment with `tech-stack.md` + project architecture; break down tasks into specific technical steps for Tier 3 Workers; maintain persistent context throughout track implementation (No Context Amnesia); review implementations + coordinate bug fixes via Tier 4 QA. CRITICAL: ATOMIC PER-TASK COMMITS (commit per task, attach git note, update plan.md). Meta-Level Sanity Check: after completing a track, run `uv run ruff check .` and `uv run mypy --explicit-package-bases .`. Anti-Entropy Protocol: state auditing before adding new state variables; TDD enforcement (Red phase before delegating). Surgical Delegation Protocol: WHERE/WHAT/HOW/SAFETY prompt template. + +**Mapping to project practice.** The Tier 2 Tech Lead role is the agent that does track execution. The skill's Atomic Per-Task Commits rule is enforced via `conductor/workflow.md` §"Task Workflow" step 9. The Meta-Level Sanity Check (ruff + mypy) is project-specific. The Anti-Entropy Protocol (state auditing + TDD enforcement) is the project's elaboration on TDD. + +**Gaps.** The Meta-Level Sanity Check rule mentions `uv run ruff check .` and `uv run mypy --explicit-package-bases .`; the project's actual command is `scripts/run_tests_batched.py` (per `conductor/workflow.md` §"Tier 2 Autonomous Sandbox"). Ruff and mypy may not be the project's primary checks. The skill's Limitations don't include "use scripts/run_tests_batched.py for tests" — this is a specific operationalization. + +**Verdict.** + +| Field | Value | +|---|---| +| **Primary** | `PARITY` | +| **Integration tag** | `INTEGRATED` | +| **Section size** | cluster | +| **Cross-refs** | n/a (project-specific) | + +**Rationale.** The skill is the Tier 2 operating procedure; followed in practice. The Meta-Level Sanity Check's specific commands may be stale. + +**Recommended change.** Update Meta-Level Sanity Check to use the project's actual check commands (`scripts/run_tests_batched.py`); add to Limitations: "Use scripts/run_tests_batched.py for verification, not raw pytest." + +### 15.4 mma-tier3-worker (Tier 3) + +**What the skill prescribes.** The `mma-tier3-worker` skill (21 lines) is the "Focused on TDD implementation, surgical code changes, and following specific specs" role. Responsibilities: implement code strictly according to the provided prompt + specifications; TDD Mandatory Enforcement (write failing test + verify it fails BEFORE writing implementation code; no `pass`-only tests; no tests without meaningful assertions; minimal + functional + conforming changes; utilize tool access). Limitations: no architectural decisions; no unrelated file modifications; stateless operation (Context Amnesia); rely on "Skeleton Views" provided by Tier 2/Orchestrator for understanding dependencies. + +**Mapping to project practice.** The Tier 3 Worker role is the stateless sub-agent that does surgical code changes. The skill is concise (21 lines) and follows the project's TDD discipline (per Section 4). The "Skeleton Views" reference maps to `manual-slop_py_get_skeleton` MCP tool. + +**Gaps.** The skill is short (21 lines) and lacks explicit reference to the OpenCode Task tool's `failure_count` escalation pattern (per `conductor/workflow.md` §"Conductor Token Firewalling" §"Handling Repeated Failures"). It also lacks explicit reference to the Pre-Delegation Checkpoint (commit before delegating per `mma-orchestrator` §"Pre-Delegation Checkpoint"). + +**Verdict.** + +| Field | Value | +|---|---| +| **Primary** | `PARITY` | +| **Integration tag** | `INTEGRATED` | +| **Section size** | cluster | +| **Cross-refs** | n/a (project-specific) | + +**Rationale.** The skill is concise and aligned with project practice. The TDD discipline is the same as Section 4's analysis. Missing references to failure_count and Pre-Delegation Checkpoint are minor. + +**Recommended change.** Add reference to OpenCode Task tool's `failure_count` escalation; add reference to Pre-Delegation Checkpoint. + +### 15.5 mma-tier4-qa (Tier 4) + +**What the skill prescribes.** The `mma-tier4-qa` skill (19 lines) is the "Focused on test analysis, error summarization, and bug reproduction" role. Responsibilities: compress large stack traces or log files into concise, actionable summaries; identify root cause of test failures or runtime errors; provide brief, technical description of the required fix; utilize diagnostic and exploration tools. Limitations: do not implement the fix directly; ensure output is extremely brief and focused; stateless operation. + +**Mapping to project practice.** The Tier 4 QA role is the stateless sub-agent that diagnoses errors. The skill is concise (19 lines) and matches the project's Tier 4 pattern (per `conductor/workflow.md` §"Conductor Token Firewalling" §"The Tier 4 QA Agent (Diagnostics)"). The "extremely brief" output constraint is project-specific. + +**Gaps.** The skill lacks explicit reference to "DO NOT make claims about success" (per the project's verification-before-completion discipline — Section 5). It also lacks explicit reference to the audit scripts available (`scripts/audit_main_thread_imports.py`, `audit_weak_types.py`, etc.) for diagnostic data. + +**Verdict.** + +| Field | Value | +|---|---| +| **Primary** | `PARITY` | +| **Integration tag** | `INTEGRATED` | +| **Section size** | cluster | +| **Cross-refs** | n/a (project-specific) | + +**Rationale.** The skill is concise and aligned with project practice. Missing references to "no success claims" and audit scripts are minor. + +**Recommended change.** Add "DO NOT make success claims; the Tier 2 reviews the fix" rule; reference audit scripts for diagnostic data. + +### 15.6 Cluster Summary + +| Sub-section | Primary | Integration tag | Recommended change | +|---|---|---|---| +| 15.1 mma-orchestrator | PARITY | INTEGRATED | Update `gemini` → OpenCode Task tool; fix typos in anti-patterns section | +| 15.2 mma-tier1-orchestrator | PARITY | INTEGRATED | Rewrite description to trigger-focused; expand Limitations | +| 15.3 mma-tier2-tech-lead | PARITY | INTEGRATED | Update Meta-Level Sanity Check commands | +| 15.4 mma-tier3-worker | PARITY | INTEGRATED | Add failure_count + Pre-Delegation Checkpoint references | +| 15.5 mma-tier4-qa | PARITY | INTEGRATED | Add "no success claims" rule; reference audit scripts | + +**Cluster-level recommendation.** All 5 MMA skills are PARITY+INTEGRATED; the cluster is well-aligned with project practice. The deferred rebuild may want to: + +- **MEDIUM:** Apply the writing-skills improvements (Section 14 recommendations) to all 5 MMA skills: trigger-focused descriptions, token-budget targets, pressure-scenario testing. +- **LOW:** Update the legacy `gemini` references and stale ruff/mypy commands. + +**Cluster verdict.** + +| Field | Value | +|---|---| +| **Primary** | `PARITY` | +| **Integration tag** | `INTEGRATED` | +| **Section size** | cluster | +| **Cross-refs** | n/a (project-specific) | + +**Rationale.** All 5 MMA skills are PARITY+INTEGRATED; the cluster faithfully represents the project's MMA pattern. The deferred rebuild can apply the writing-skills improvements to tighten each skill. ## 16. Dual-Convention + Anything Else