feat(reusable): ship the curated swarm as the default config consumers inherit #10

Merged
steve merged 4 commits from feat/default-swarm-config into main 2026-06-28 02:23:40 +00:00
6 changed files with 64 additions and 40 deletions
+10 -13
View File
@@ -1,10 +1,11 @@
# Gadfly reviewing its OWN PRs — now a thin CALLER of the reusable workflow
# (.gitea/workflows/review-reusable.yml), dogfooding the Phase-4 "subscribe"
# path. The reusable holds the image pin + env plumbing; this file holds only
# the triggers, the actor gate, and gadfly's specific swarm config.
# Gadfly reviewing its OWN PRs — a thin CALLER of the reusable workflow
# (.gitea/workflows/review-reusable.yml), dogfooding the "subscribe" path. The
# reusable holds the image pin, env plumbing, AND the default swarm; this file
# holds only the triggers, the actor gate, secret forwarding, and allow-list.
#
# Advisory only — never blocks a merge. Fleet: 6 cloud + Claude Code
# (sonnet, opus, opus:max) competitors.
# Advisory only — never blocks a merge. It inherits the default swarm: 3 cloud
# models + Claude Code (sonnet, opus, opus:max), 5-lens suite (claude models run
# one at a time, each with all 5 lenses at once).
name: Adversarial Review (Gadfly)
@@ -48,11 +49,7 @@ jobs:
GADFLY_FINDINGS_URL: ${{ secrets.GADFLY_FINDINGS_URL }}
GADFLY_FINDINGS_TOKEN: ${{ secrets.GADFLY_FINDINGS_TOKEN }}
with:
models: "minimax-m3:cloud,glm-5.2:cloud,glm-5.1:cloud,deepseek-v4-pro:cloud,nemotron-3-super:cloud,qwen3-coder:480b-cloud,claude-code/sonnet,claude-code/opus,claude-code/opus:max"
specialists: "security,correctness,error-handling"
provider_concurrency: "ollama-cloud=3,claude-code=2"
provider_lens_concurrency: "ollama-cloud=3"
timeout_secs: "600"
max_steps: "14"
# Inherit the default swarm (3 cloud + Claude Code sonnet/opus/opus:max,
# 5-lens suite) from review-reusable.yml. Only the consumer-specific
# allow-list is set here.
allowed_users: "steve,fizi,dazed"
timeout_minutes: 90
+28 -14
View File
@@ -10,11 +10,13 @@
# secrets: # forward ONLY what the reviewer needs
# OLLAMA_CLOUD_API_KEY: ${{ secrets.OLLAMA_CLOUD_API_KEY }}
# CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
# with: { models: "...", allowed_users: "..." } # all optional
# with: { allowed_users: "..." } # config inputs are optional (see below)
#
# Inputs are all optional and default to "" — an empty env value makes the
# image/entrypoint use its own built-in default, so the caller only sets what it
# wants to override. Secrets are DECLARED below (workflow_call.secrets) so a
# Inputs ship the DEFAULT swarm (see the inputs block): 3 cloud models + the
# Claude Code engine, 5-lens suite (claude models serial, 5 lenses each). A consumer
# inherits it by omitting `with:` entirely, or overrides any field (e.g.
# `models:` for a cloud-only / different-provider setup; "" falls back to the
# image's built-in default). Secrets are DECLARED below (workflow_call.secrets) so a
# caller forwards only the credentials the reviewer actually uses — least
# privilege — rather than `secrets: inherit`, which leaks every caller secret
# (registry/deploy/db creds) into this workflow. `secrets: inherit` still works
@@ -30,21 +32,33 @@ name: Gadfly review (reusable)
on:
workflow_call:
# Inputs ship the DEFAULT Gadfly swarm so a consumer can just call this
# workflow (no `with:` block) and inherit it. The default is opinionated —
# 3 strong cloud models + the Claude Code engine (sonnet/opus/opus:max), the
# 5-lens suite. Claude models run ONE at a time, but each runs all 5 lenses at
# once. It needs OLLAMA_CLOUD_API_KEY and CLAUDE_CODE_OAUTH_TOKEN; a consumer
# with only one (or a different provider) overrides `models:` (and forwards
# just the secrets it uses). Set any input to "" to fall back to the
# image/entrypoint built-in default.
#
# Peak claude concurrency = provider_concurrency × provider_lens_concurrency
# (1 model × 5 lenses = up to 5 concurrent `claude -p` per pass) — friendly to
# a single subscription. Raise claude-code in either knob for more parallelism.
inputs:
models: { type: string, default: "" } # GADFLY_MODELS (csv)
specialists: { type: string, default: "" } # GADFLY_SPECIALISTS
models: { type: string, default: "minimax-m3:cloud,glm-5.2:cloud,deepseek-v4-pro:cloud,claude-code/sonnet,claude-code/opus,claude-code/opus:max" } # GADFLY_MODELS (csv)
specialists: { type: string, default: "security,correctness,maintainability,performance,error-handling" } # GADFLY_SPECIALISTS (5-lens default suite)
provider: { type: string, default: "" } # GADFLY_PROVIDER
base_url: { type: string, default: "" } # GADFLY_BASE_URL
provider_concurrency: { type: string, default: "" } # GADFLY_PROVIDER_CONCURRENCY
provider_lens_concurrency: { type: string, default: "" } # GADFLY_PROVIDER_LENS_CONCURRENCY
timeout_secs: { type: string, default: "" } # GADFLY_TIMEOUT_SECS (per lens)
max_steps: { type: string, default: "" } # GADFLY_MAX_STEPS
provider_concurrency: { type: string, default: "ollama-cloud=3,claude-code=1" } # GADFLY_PROVIDER_CONCURRENCY (claude models one at a time)
provider_lens_concurrency: { type: string, default: "ollama-cloud=3,claude-code=5" } # GADFLY_PROVIDER_LENS_CONCURRENCY (each claude runs all 5 lenses at once)
timeout_secs: { type: string, default: "600" } # GADFLY_TIMEOUT_SECS (per lens)
max_steps: { type: string, default: "14" } # GADFLY_MAX_STEPS
worker_model: { type: string, default: "" } # GADFLY_WORKER_MODEL
allowed_users: { type: string, default: "" } # GADFLY_ALLOWED_USERS
allowed_users: { type: string, default: "" } # GADFLY_ALLOWED_USERS (consumer-specific; set in your stub)
trigger_phrase: { type: string, default: "" } # GADFLY_TRIGGER_PHRASE
# Job wall-clock cap. 45 > 30 as a default: a multi-model swarm or a slow
# lens (e.g. claude-code with extended thinking) can exceed 30 minutes.
timeout_minutes: { type: number, default: 45 }
# Job wall-clock cap. 90 as a default: the 5-lens suite across a slow lane
# (claude-code with extended thinking) over two passes can run long.
timeout_minutes: { type: number, default: 90 }
# Declared so callers can forward ONLY the secrets the reviewer needs
# (least privilege) instead of `secrets: inherit`, which would hand this
# workflow every secret in the caller's repo (registry/deploy/db creds the
+9 -4
View File
@@ -46,9 +46,11 @@ entrypoint.sh container brains: trigger gating, PR clone, model loop (t
used to live in workflow YAML)
Dockerfile multi-stage; private-module creds via BuildKit secrets never reach the final image
.gitea/workflows/build-image.yml push main → :latest; tag v* → :<tag>+:latest; PR → build-only
.gitea/workflows/review-reusable.yml reusable (workflow_call) review job; consumers subscribe with
an ~8-line caller forwarding only the secrets the reviewer needs (Phase 4). gadfly's own
adversarial-review.yml is a thin caller of it (dogfoods the path).
.gitea/workflows/review-reusable.yml reusable (workflow_call) review job; ships the DEFAULT swarm as
input defaults (3 cloud + Claude Code sonnet/opus/opus:max, 5-lens suite;
claude models serial, 5 lenses each) so consumers inherit it by omitting `with:`. Consumers subscribe
with an ~8-line caller forwarding only the secrets the reviewer needs (Phase 4);
gadfly's own adversarial-review.yml is a thin caller of it (dogfoods the path).
examples/ copy-paste consumer stub workflows for different providers
```
@@ -98,7 +100,10 @@ comment with a section each. Default suite = security/correctness/maintainabilit
error-handling; opt-in built-ins = tests/docs/conventions/improvements. Select via
`GADFLY_SPECIALISTS` (csv or `all`); define/override via `GADFLY_SPECIALIST_<NAME>` env or a repo
`.gadfly.yml` (`specialists:` + `define:`). See `cmd/gadfly/specialists.go`. Cost ≈
specialists × models × 2 passes — keep the default model count low (entrypoint defaults to one).
specialists × models × 2 passes — the **image/entrypoint** default stays minimal (one model) for
that reason; the **reusable workflow** (`review-reusable.yml`) deliberately ships a heavier
opinionated default swarm (3 cloud + Claude Code, 5 lenses) for steve's own fleet, which consumers
inherit or override per-input.
**Dynamic `auto`** (`GADFLY_SPECIALISTS=auto`): a selector (`GADFLY_SELECTOR_MODEL` or the review
model) picks lenses per-diff and may invent ad-hoc ones (`cmd/gadfly/auto.go`). **Worker-tier**
(`GADFLY_WORKER_MODEL`): a `delegate_investigation` tool offloads grep/read legwork to a cheap
+2 -1
View File
@@ -39,7 +39,8 @@ it. Drop one file in your repo and set a couple of secrets/vars:
1. Copy a stub from [`examples/`](examples/) to `.gitea/workflows/adversarial-review.yml` in
your repo. Two flavors: the slim [`reusable.yml`](examples/reusable.yml) — a tiny caller of
Gadfly's **reusable workflow** (`uses: steve/gadfly/.gitea/workflows/review-reusable.yml@…`,
forwarding only the secrets the reviewer needs), best when you take the defaults — or the full self-contained
forwarding only the secrets the reviewer needs), which ships a **default swarm** (3 cloud models +
the Claude Code engine, 5-lens suite) you inherit by omitting `with:` or override per-input — or the full self-contained
[`adversarial-review.yml`](examples/adversarial-review.yml) (Ollama Cloud default, with inline
notes for every provider / local Ollama / OpenAI-compatible / endpoint aliases). See the
[examples index](examples/README.md).
+1 -1
View File
@@ -6,7 +6,7 @@ set the secrets/vars it references. Gadfly is advisory only — it never blocks
| File | Backend | Needs |
|------|---------|-------|
| [`reusable.yml`](reusable.yml) | **slimmest stub** — calls Gadfly's reusable workflow, forwarding only the secrets the reviewer needs (least privilege, not `secrets: inherit`); take the defaults or override a few inputs | secret `OLLAMA_CLOUD_API_KEY` |
| [`reusable.yml`](reusable.yml) | **slimmest stub** — calls Gadfly's reusable workflow and inherits its **default swarm** (3 cloud + Claude Code, 5-lens suite), forwarding only the secrets it needs (least privilege, not `secrets: inherit`); the stub keeps a cloud-only `models:` override so it runs with just the Ollama key (drop it + add the Claude token to get the full default) | secret `OLLAMA_CLOUD_API_KEY` |
| [`adversarial-review.yml`](adversarial-review.yml) | **Ollama Cloud** (default) + inline notes for every provider; full self-contained stub | secret `OLLAMA_CLOUD_API_KEY` |
| [`local-ollama.yml`](local-ollama.yml) | a **local/LAN Ollama** daemon | nothing (or `GADFLY_BASE_URL` for a remote host) |
| [`openai-compatible.yml`](openai-compatible.yml) | any **OpenAI-compatible** endpoint (local Ollama `/v1`, gateway, vLLM, OpenRouter…) | `GADFLY_BASE_URL` (+ a key for most gateways) |
+14 -7
View File
@@ -5,7 +5,12 @@
# workflow, which holds the image pin + all the env plumbing. You only declare
# the triggers, the comment-trigger actor gate, and any overrides you want.
#
# Needs: secret OLLAMA_CLOUD_API_KEY (the default Ollama Cloud provider).
# The reusable ships a DEFAULT swarm: 3 cloud models + the Claude Code engine
# (sonnet/opus/opus:max), 5-lens suite. That default needs BOTH
# OLLAMA_CLOUD_API_KEY and CLAUDE_CODE_OAUTH_TOKEN. This example overrides
# `models:` to a cloud-only set so it works with just OLLAMA_CLOUD_API_KEY —
# delete that override (and forward the Claude token) to inherit the full default.
#
# Forward ONLY the secrets the reviewer uses (least privilege) — see the
# `secrets:` block below. GITEA_TOKEN is automatic. `secrets: inherit` also works
# but hands the reusable EVERY secret in your repo (registry/deploy/db creds the
@@ -13,9 +18,8 @@
# Gadfly commit (not @main) so a push there can't change what runs with your
# secrets.
#
# Prefer this when you're happy with the defaults. For custom named endpoints
# (GADFLY_ENDPOINT_<NAME>) or a provider the reusable doesn't map, use the full
# stub in adversarial-review.yml instead.
# For custom named endpoints (GADFLY_ENDPOINT_<NAME>) or a provider the reusable
# doesn't map, use the full stub in adversarial-review.yml instead.
name: Adversarial Review (Gadfly)
@@ -56,7 +60,10 @@ jobs:
# GADFLY_FINDINGS_URL: ${{ secrets.GADFLY_FINDINGS_URL }}
# GADFLY_FINDINGS_TOKEN: ${{ secrets.GADFLY_FINDINGS_TOKEN }}
with:
# All optional — omit to take Gadfly's defaults. Examples:
# models: "qwen3-coder:480b-cloud,gpt-oss:120b-cloud"
# specialists: "security,correctness,error-handling"
# Cloud-only override so this works with just OLLAMA_CLOUD_API_KEY. Delete
# this line (and forward CLAUDE_CODE_OAUTH_TOKEN above) to inherit the full
# default swarm (3 cloud + Claude Code sonnet/opus/opus:max, 5 lenses).
models: "minimax-m3:cloud,glm-5.2:cloud,deepseek-v4-pro:cloud"
# Other inputs inherit the default (5-lens suite, concurrency, 90-min cap);
# override any of them here (specialists, provider, base_url, timeout_secs…).
allowed_users: "your-username"