Skip to content

pipeline.md

Latest commit

 

History

History
233 lines (175 loc) · 8.56 KB

pipeline.md

File metadata and controls

233 lines (175 loc) · 8.56 KB
Raw

🏭 Pipeline config

The full engine graph — phases, loop agents, and insertable slots — is data-driven. Studio, the progress header, and the orchestrator all read the same file:

.slmcode/pipeline.yaml

🧩

Insert any agent anywhere: before/after/replace a phase, pick a custom or built-in agent, control the prompt template, fail mode, and where output lands. The UI follows the config.

Quick start

slmcode init                 # writes default pipeline.yaml
slmcode blocks apply go      # apply Go-optimized pipeline preset
slmcode studio               # Pipeline tab → edit → Save
curl -s localhost:7420/api/pipeline | jq .

Or switch presets from the Studio:

  • Pipeline tab: Use the preset selector (every pipeline block the registry sees) for one-click switching
  • Blocks tab: Browse all pipeline presets and apply any
  • Settings: Use the Pack Selector to switch the entire language workflow

Reset to built-ins:

curl -s -X POST localhost:7420/api/pipeline/reset
# or via CLI:
slmcode blocks apply go   # re-apply any preset

---

## Document shape

```yaml
version: 1
order: [init, skills, context, explore, docs, architect, clarify, plan, split, coord, execute, learn, polish, test, memory, done]

groups:
  - id: prepare
    label: Prepare
    steps: [init, skills, context, explore, docs]
  # …

phases:
  context:
    agent: context
    when: always          # always | auto | never
    label: Context
    tip: Refresh CONTEXT
    group: prepare
  explore:
    agent: explorer
    when: auto
  plan:
    agent: planner
    when: always
  test:
    agent: tester
    when: always

execute:
  default_role: worker
  reviewer: reviewer      # any registered agent id (custom OK)
  corrector: corrector
  max_waves: 2

slots:
  - id: pre-plan-audit
    agent: night-auditor  # custom or builtin
    after: explore        # or before: plan  or replace: plan
    when: always          # or never | query_matches:langgraph
    input: |
      Audit exploration before planning.
      Query:
      {{query}}
      Exploration:
      {{exploration}}
    persist_to: scratch   # scratch | context | memory | none
    fail_mode: continue   # continue | abort
    multipass: false

Phase when

Value Behavior
always Run every time
auto Built-in heuristics (explore/architect/docs)
never Skip (or use a replace slot instead)

Slot placement

Exactly one of:

  • after: <phase> — run after the phase finishes
  • before: <phase> — run before the phase
  • replace: <phase> — run instead of the built-in phase agent

Prompt placeholders

{{query}} · {{exploration}} · {{plan}} · {{phase}}

API

Method Path Purpose
GET /api/pipeline Resolved config + anchors + defaults
PUT /api/pipeline Body { "config": { … } }
POST /api/pipeline/reset Restore defaults

Agents themselves stay under /api/agents (prompt, model, tools, …). The pipeline only references agent IDs and placement.

Studio

Pipeline tab:

  1. Execute loop — pick reviewer / corrector / default worker
  2. Phase agents — bind any registered agent per stage + when
  3. Slots — insert agents around phases with a prompt template

The top progress header is driven by the same order / groups / slots — no hard-coded stage list in the UI.

Board task role dropdowns also list all agents from /api/agents (customs included).

Custom agents + pipeline

  1. Create @night-auditor in Agents (or .slmcode/agents/night-auditor.yaml)
  2. Open Pipeline → add slot after: explore → agent night-auditor
  3. Save → next full run executes the slot and streams it live

Specialist mode also accepts custom agent IDs.

Predefined pipeline presets

SLMCode ships with thirteen built-in pipeline presets, each optimized for a specific language:

Preset Language Tester Agent Worker Agent QA Gate
go 🐹 Go go-tester go-worker go test ./... -race -count=1
python 🐍 Python python-tester python-worker python -m pytest -q
react ⚛️ React react-tester react-worker npm test --silent
typescript 🟦 TypeScript / Node ts-tester ts-worker npm test --silent
web 🌐 Static HTML/CSS/JS web-tester web-worker non-empty index.html entrypoint
rust 🦀 Rust rust-tester rust-worker cargo test --quiet
java ☕ Java java-tester java-worker mvn -q -B test
kotlin 🟪 Kotlin kotlin-tester kotlin-worker ./gradlew test --console=plain
dotnet 🟣 C# / .NET dotnet-tester dotnet-worker dotnet test --nologo --verbosity quiet
ruby 💎 Ruby ruby-tester ruby-worker bundle exec rspec --no-color
php 🐘 PHP php-tester php-worker vendor/bin/phpunit --colors=never
swift 🕊️ Swift swift-tester swift-worker swift test
cpp ⚙️ C/C++ cpp-tester cpp-worker ctest --test-dir build --output-on-failure

Each preset:

  • Sets the test phase agent to a language-specific verifier
  • Configures the execute loop with a language-aware worker
  • Adds quality gate slots with language-specific check reminders
  • Pins relevant skills like atomic-coding and specialist-tester

Apply any preset via CLI, API, or Studio:

# CLI
slmcode blocks apply python

# API
curl -X POST localhost:7420/api/packs/python/apply

# API — apply just the pipeline (no QA gate)
curl -X POST localhost:7420/api/pipeline-presets/python/apply

Custom presets can be created as YAML blocks — see 🧱 Blocks.

What a run skips, and why

A phase that is enabled still only runs when it has something to do. Every skip is announced as an info event in the phase's own lane, so the Live view and the CLI show the reason rather than a phase that silently did not happen.

Phase Runs when Skipped with
split (the splitter) the plan has two or more steps, or one step over more than two known files, or no known file at all splitter skipped — a 1-step plan over N known file(s) is one task; building the board directly. The board is built from the plan's one step with the known files as its scope.
per-wave coordinator (coord @after-wave) the wave that just finished failed a task, escalated one, or produced a lesson worth keeping — a failure lesson or an honored human note; the routine "this task passed its acceptance" lessons every green task yields do not count coordinator @after-wave skipped — all N task(s) finished green with no failure, escalation or new lesson
per-wave distillation (learn, think_passes >= 2) same rule as the coordinator wave distillation skipped — …
memory (end-of-run distillation) the run left a lesson worth keeping (same notion), changed a file, or failed a task memory distillation skipped — nothing to distill: no lessons, no changed files, no failed tasks. The distiller is a single call now, not a multipass cycle.

Two more things the pipeline no longer repeats:

  • One concurrency limit for the whole run. max_parallel used to bound only the execute wave; context ran beside explore, architect beside clarify, speculative digs and review races added slots on top. Every model request — phase roles, workers, reviewers, correctors, critique, triage, the planner's multipass cycle — now takes a slot of one run-wide gate sized max_parallel, and at max_parallel: 1 the phase pairs run one after another.
  • The objective command runs once per tree state. The deterministic pre-test, each team's acceptance, the integration command and the QA gate's first round all ask the same question of the same tree; a result is remembered per (command, tree fingerprint) and reused until something is written — an agent write, a rewrite of an already-changed file, a formatter pass or a dependency install all count.

Related

☀️ Made with ♥ by UnicoLab