agentsclimarketplace

Qrspi workflow

Skill bostonaholic/team/skills/qrspi-workflow

Worktree-Question-Research-Design-Structure-Plan-Implement-PR phase discipline with gate enforcement — loaded by orchestrator to govern pipeline phase transitions, artifact conventions, and anti-patternsFrom its SKILL.md

Install
npx -y skills add bostonaholic/team --skill qrspi-workflow

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 11 stars11 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

8.2 KB, as published. Nobody here has run it

QRSPI Workflow

Phase discipline for the Team pipeline. Eight sequential phases, none skippable. Each produces artifacts the next consumes.

Phase Sequence

WORKTREE -> QUESTION -> RESEARCH -> DESIGN -> STRUCTURE -> PLAN -> IMPLEMENT -> PR
PhaseProducesGate
WORKTREEgit worktree on branch <id> off origin/HEAD, with docs/plans/<id>/ authored inside itHARD — the worktree must exist before QUESTION authors artifacts
QUESTIONtask.md (full description, human-only) and questions.md (neutral questions plus a "Codebase context" section naming files, modules, and vocabulary but NOT the goal)HARD — both on disk
RESEARCHresearch.mdHARD — artifact on disk
DESIGNdesign.md (~200 lines) plus one design-review-<n>.md per roundREVIEW — adversarial design review. APPROVE/COMMENT advance; REQUEST CHANGES re-drafts, cap 5 → terminal halt
STRUCTUREstructure.md (~2 pages of vertical slices)NONE — autonomous
PLANplan.mdSOFT — no approval. The reviewed design is the contract
IMPLEMENTproduction code, passing tests, per-slice commitsAGGREGATE — security, verifier, and code-review hard gates
PRGitHub draft PRTerminal — record the PR URL, close the ledger

WORKTREE is the leading phase and has no agent — purely a router responsibility. For why it leads, see "Why first" in skills/worktree-isolation/SKILL.md.

IMPLEMENT runs four sub-phases: test-architect writes failing acceptance tests; a mechanical gate confirms all tests fail with assertion errors (not crashes) and every static check the project defines passes; implementer works through vertical slices, committing each when its tests pass; then 5 parallel reviewers (code, security, docs, ux, verifier) return typed failure classes that loop back to the implementer, capped at 5 rounds.

Artifact Conventions

All phase artifacts live under docs/plans/<id>/. The schema is canonical in skills/artifact-frontmatter/SKILL.md — the <id> forms, the artifact inventory, the repos.md and prd.md schemas, the topic-consistency invariant, and the ticketId scope. Consult that skill rather than restating it. What matters for phase discipline:

  • The <id> slug and the topic frontmatter field match across every artifact for the same feature.
  • repos.md (when present) switches the pipeline into multi-repo mode. Its absence keeps single-repo mode — today's default.
  • prd.md (when present) rides the autonomous Question phase and is not gated.

Research Isolation

Research is the most-corruptible phase: a model that knows what it is being asked to build returns opinions instead of facts. The invariant is enforced in two layers:

  1. Structural — the orchestrator passes researcher and file-finder only the path to questions.md. It is forbidden from handing them the description or task.md at dispatch time.
  2. Procedural — the researcher and file-finder system prompts forbid reading task.md. Both hold Read/Grep/Glob with permissionMode: plan, so nothing mechanically stops such a read. Enforcement relies on the agent following its prompt. A researcher missing context surfaces it as an open question rather than guessing the intent, and never pauses the run to ask.

Gate Types

HARD

Blocks the phase transition until satisfied. No override except by explicit user command. Examples: a REQUEST CHANGES design-review verdict, security findings that gate, test failures.

SOFT

Informational. SOFT findings land in the PR body's ## Review notes for the human's PR review and are never acknowledged mid-run. The pipeline proceeds.

Which findings gate, and which auto-fix rather than land as recorded notes, is defined in exactly one place: skills/review-severity-tiers/SKILL.md → "Severity Tiers and the Auto-Fix Boundary". Only findings below the auto-fix boundary are recorded for the PR body. Consult that table rather than restating it here.

ADVISORY

Non-blocking, no acknowledgment required. Examples: documentation gap analysis, style suggestions.

State and Coordination

Pipeline state is reconstructed by scanning docs/plans/<id>/*.md and reading YAML frontmatter. The orchestrator tracks in-flight work through TodoWrite — a session-scoped ledger mirroring the phase table, rebuilt on entry to any /team-* command.

Phase inference from artifacts

Latest artifact presentCurrent phase
worktree exists for <id>, no task.md yetWORKTREE (next up)
task.md + questions.mdRESEARCH (next up)
research.mdDESIGN (next up)
design.md alone (no passing design review)DESIGN (review next)
design.md + passing design-review-<n>.mdSTRUCTURE (next up)
structure.mdPLAN (next up)
plan.md + ≥1 commit on <id> since merge-baseIMPLEMENT
plan.md (no commit on <id> yet)PLAN (next up)
topic branch has commits ahead and verifier passedPR (next up)
PR(s) opened or commit(s) shippedSHIPPED

Worktree presence (single-repo): git worktree list --porcelain | grep -q <id>. Multi-repo: the same per repo path in docs/plans/<id>/repos.md. IMPLEMENT signal: a worktree alone is not enough — IMPLEMENT is confirmed only once git log <merge-base>..<id> is non-empty. Before that, plan.md present with no commit means the run is still pre-IMPLEMENT. Verifier passed: the latest docs/plans/<id>/review-<n>.md shows the aggregate gate clean.

Phase Transition Protocol

  1. Verify artifacts exist on disk for the current phase. For DESIGN → STRUCTURE that includes a design-review-<n>.md with a passing verdict.
  2. Update the ledger — current TodoWrite item complete, next in_progress.
  3. Dispatch next agent(s) — the phase table in skills/team/SKILL.md names them.

Never proceed while a Blocking or Major finding remains. The implementer loops automatically and the user is never consulted about it — the no-consult rule in skills/review-severity-tiers/SKILL.md. Minor-and-below findings are recorded for the PR body's ## Review notes, never presented mid-run.

Anti-Patterns

  • Skipping Question. The researcher then inherits the user's framing and produces opinionated findings. Always run the questioner first.
  • Letting research see intent. A researcher that reads task.md or receives the description in any form breaks the isolation invariant. Treat any leakage as a critical defect: stop and report.
  • Reviewing the plan. A 1000-line plan begets ~1000 lines of code, and surprises during implementation invalidate the review. Review the design (~200 lines) instead — that is where leverage lives. The structure and plan are autonomous tactical artifacts.
  • Horizontal layering. A structure that builds the entire database, then the entire API, then the entire UI defers integration risk to the end. Reject any structure that flattens into layers — slices are end-to-end, independently testable, and atomically committable (skills/slicing-work/SKILL.md).
  • Implementing without a structure. The structure is the scope fence. Always produce it, even though it advances autonomously.
  • Gold-plating. Adding features, tests, or abstractions beyond what the structure specifies. If scope must expand, update the structure; for a material change, return to DESIGN for a fresh design review.
  • Backward skipping. Jump back one phase, never several. A structure flaw returns to STRUCTURE; a design flaw returns to DESIGN.
  • Premature shipping. Every HARD gate in the implement-verify loop must pass before the PR phase.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most agent orchestration skills give

Counted across 848 of the 1,300 authors here whose files we hold, read 2026-09-06

  • Dispatch one agent per independent problem domainin 56 of 848, across 42 files
  • Run full test suite after integrationin 55 of 848, across 42 files
  • Verify fixes do not conflictin 40 of 848, across 32 files
  • Review each summary when agents returnin 40 of 848, across 31 files
  • Write a handoff document summarising the current conversationin 30 of 848, across 25 files
  • Reference existing artifacts by path or URLin 26 of 848, across 24 files
  • Give each agent a specific scopein 19 of 848, across 10 files
  • Give each agent a clear goalin 19 of 848, across 10 files
  • Include a suggested skills section in the documentin 18 of 848, across 16 files
  • Tailor the doc to the user argumentsin 18 of 848, across 15 files
  • Issue all subagent dispatches in the same responsein 17 of 848, across 11 files
  • Use git worktrees for isolationin 17 of 848, across 8 files

Said here and by no other author read

  • Execute eight sequential phases in strict order
  • Store all phase artifacts under the plans directory
  • Match the id slug and topic across every artifact
  • Pass only the questions path to researcher and file finder
  • Reconstruct pipeline state by scanning markdown artifacts
  • Never proceed while a blocking or major finding remains

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.

Keep looking

Skills are one crate of 611,676. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.