internal/infrastructure/configuration/setup.Run drives operator identity, LLM provider, models, forge, and Telegram chat channel prompts and returns TOML edits -- not a config.Config. Building a parallel in-memory representation to validate against was rejected during design review: it's a second source of truth about the exact thing (text vs. struct disagreeing) this whole feature exists to stop. Tests instead render the edits against the real embedded template and load the result through a real configuration.Loader, proving the actual defaults/decode/validate path accepts it. Two boot-time landmines in config.example.toml surfaced by that approach, both fixed here: - [providers.openai] ships active with api_key pointed at the bws engine. cmd/archied/provider_secrets.go resolves every entry in cfg.Providers at startup, not just the ones [models] references, and refuses to start the daemon if any fails -- so picking any other provider (or picking openai but skipping the key) left a config that wouldn't boot on a machine without the bws CLI installed (the common case). stepProvider now neutralises providers.openai.api_key to the zero-value secret ref whenever openai wasn't the actual choice. - [[repos]] shipped active with a placeholder owner/name that validates cleanly (owner/name are only checked non-empty) and would have archied silently polling a repo that doesn't exist. Commented out in the template with an explanation; setup does not write this section since tomlwrite cannot target array-of-tables. TestExampleConfigLoads updated to expect zero repos. Also: forge and provider secret env-var names are each defined in one table (forgeToken, cloudProviders) read by both the config edit and the secret sink, so the two can't name a secret differently -- the exact bug class this feature exists to close for the installer, tested directly (TestRun_ForgeTokenKeyMatchesStoredSecretKey compares the decoded key against the key the secret was actually stored under, not a hardcoded literal on both sides). Telegram writes chat.telegram. token_env, not a secret.SecretRef -- TelegramConfig has no such field. Telegram also requires at least one allowed_user_id: AllowedUserIDs is deny-by-default, so a bot token with none configured answers nobody, which reads as broken rather than as the safe default it is. SecretSink is a caller-supplied sink (Put buffers, Commit persists), not a return value: Run never hands back plaintext secrets, and the caller is expected to call Commit only after proving the generated config loads -- so a validation failure never leaves a secret written without the config that references it. Refs archie-core-cbk
160 lines
6.7 KiB
TOML
160 lines
6.7 KiB
TOML
# archied configuration. Copy to ~/.config/archie/config.toml.
|
|
# Secrets are NOT configured here directly: each one is a {engine, key}
|
|
# reference resolved through a secrets engine at startup. The built-in
|
|
# "env" engine reads key as an environment variable name. Custom engines
|
|
# (sops, bws, vault/openbao, ...) are Yaegi plugins loaded from a
|
|
# directory -- see internal/secret.
|
|
# secret_engine_dir = "~/.config/archie/secret-engines"
|
|
|
|
# work_dir = "~/.local/share/archie/work" # default
|
|
# db_path = "~/.local/share/archie/archie.db"
|
|
poll_interval = "60s"
|
|
label = "archie"
|
|
bot_user = "vectis-labs-archie"
|
|
diff_cap_lines = 400
|
|
# max_retries = 3 # how many operator-triggered retries a parked task gets
|
|
# before it moves to dead (global default; set per-repo to
|
|
# override). Enforced by the dashboard's retry action.
|
|
|
|
[forge]
|
|
type = "github"
|
|
host = "https://github.com"
|
|
token = { engine = "env", key = "ARCHIE_GITHUB_TOKEN" }
|
|
|
|
[dispatch]
|
|
trigger = "assignee" # "assignee", "label", or "either"
|
|
ack_reaction = "eyes" # "off" disables pickup reactions
|
|
|
|
[dispatch.labels]
|
|
queued = "archie:queued"
|
|
working = "archie:working"
|
|
waiting = "archie:waiting"
|
|
pr = "archie:pr"
|
|
parked = "archie:parked"
|
|
# dead = "archie:dead" # state label for a task that exhausted max_retries
|
|
|
|
[models] # role -> ai-sdk runtime ref (used by agent workflows, Phase 4b)
|
|
triage = "openai/gpt-5.4"
|
|
planner = "openai/gpt-5.4"
|
|
builder = "openai/gpt-5.4"
|
|
|
|
[providers.openai]
|
|
class = "openai"
|
|
api_key = { engine = "bws", key = "OPENAI_API_KEY" }
|
|
# api_key_env = "OPENAI_API_KEY" # backwards-compatible environment form
|
|
|
|
[providers.ollama]
|
|
class = "ollama"
|
|
|
|
# Optional interactive-chat catalog. /provider groups these references by
|
|
# their first path segment, and /model shows only the selected provider's
|
|
# entries. When omitted, Archie uses the distinct refs in [models].
|
|
#
|
|
# `operator` is the display name of the person this deployment assists. It is
|
|
# passed to the chat agent as runtime metadata, which means it is sent to the
|
|
# configured model provider on every turn -- leave it unset to tell the agent
|
|
# nothing about who is running it.
|
|
#
|
|
# `workspace` roots the chat agent's file and shell tools (read, write, edit,
|
|
# shell, grep, find). Unset, those tools are not registered at all. Set it and
|
|
# the agent can read, modify and execute inside that directory with the same
|
|
# reach the daemon process has -- choose it deliberately.
|
|
#
|
|
# `max_steps` caps the model/tool round-trips in a single chat turn
|
|
# (default 100). This is a task budget, not a safety limit: one question
|
|
# about a codebase easily costs a read, several greps and a few edits, and
|
|
# a turn that runs out stops mid-task looking like it gave up. To interrupt
|
|
# a turn, send /stop -- that cancels it immediately and clears anything
|
|
# queued behind it.
|
|
# [chat]
|
|
# operator = "Ada Lovelace"
|
|
# workspace = "~/.local/share/archie/workspace"
|
|
# max_steps = 100
|
|
# models = [
|
|
# "openai/gpt-5.6-sol",
|
|
# "openrouter/openai/gpt-5.6-sol",
|
|
# ]
|
|
|
|
# Optional self-hosted update adapter. Both values are argv arrays (never
|
|
# shell snippets). The check command must write a releaseupdate.Snapshot JSON
|
|
# document to stdout; configure the installer as well to show Approve/Defer.
|
|
# [chat.telegram]
|
|
# update_check_command = ["/usr/local/bin/archie-update-check"]
|
|
# update_install_command = ["/usr/local/bin/archie-update-install"]
|
|
|
|
# External MCP tools are launched and managed by Archie's MCP client.
|
|
# [[tools.mcp_servers]]
|
|
# name = "desktop-commander"
|
|
# transport = "stdio"
|
|
# command = "npx"
|
|
# args = ["-y", "@wonderwhy-er/desktop-commander@0.2.46", "--no-onboarding"]
|
|
# work_dir = "/path/to/workspace"
|
|
|
|
[agent]
|
|
mode = "inprocess" # "subprocess" runs each stage through archie-agent (stdin/stdout); "nats" sends stages over NATS
|
|
command = "archie-agent" # executable path for subprocess mode (not used in nats mode)
|
|
# Subprocess mode isolates the protocol and process lifecycle, not OS privileges.
|
|
# The worker still runs as the daemon user until the Docker sandbox phase lands.
|
|
# env = ["GOCACHE"] # extra environment variable names forwarded to the worker
|
|
|
|
[indexing] # workspace codesearch index; both paths default under work_dir
|
|
# index_dir = "/var/lib/archie/indexes" # codesearch sidecar files
|
|
# db_path = "/var/lib/archie/workspace-indexes.db" # index lifecycle metadata
|
|
|
|
[web] # observability dashboard (read-only, no auth -- keep it local/tailnet)
|
|
listen = "127.0.0.1:8484" # "off" disables
|
|
|
|
[notify] # optional webhook for human-in-the-loop moments (n8n -> email)
|
|
# webhook = "https://n8n.example.com/webhook/archie"
|
|
|
|
[budgets]
|
|
max_steps = 60
|
|
max_tokens = 400000
|
|
wall_clock = "30m"
|
|
gate_max_failures = 5
|
|
|
|
# NATS JetStream for distributed task queue. Remove or leave commented out
|
|
# to run in single-process SQLite mode.
|
|
# [nats]
|
|
# url = "nats://localhost:4222"
|
|
# token_env = "NATS_TOKEN" # optional; named variable must be set
|
|
|
|
# Docker sandbox execution of archie-agent. Requires [nats] and agent.mode = "nats".
|
|
# [containers]
|
|
# enabled = true
|
|
# image = "archie-agent:latest" # Docker image to run
|
|
# max_concurrency = 4 # max simultaneous tasks/containers
|
|
# max_uptime = "60m" # container lifetime cap
|
|
# volume_ttl = "72h" # max age for persistent repo volumes and Git clone caches
|
|
# pull_policy = "missing" # "missing" or "always"
|
|
|
|
# At least one [[repos]] entry is required before archied will poll
|
|
# anything -- it is deliberately left commented rather than shipped active:
|
|
# an active example here validates as a normal, working repo (owner/name are
|
|
# only checked for non-empty), so archied would silently start polling a
|
|
# repository that doesn't exist. archied setup does not write this section
|
|
# ([[repos]] is an array of tables, which the text-patching setup uses
|
|
# cannot target) -- add at least one entry by hand.
|
|
#
|
|
# [[repos]]
|
|
# owner = "your-github-username"
|
|
# name = "todo"
|
|
# base = "main"
|
|
# ecosystem = "go" # "go", "python", "node", "rust", or "custom"
|
|
# # Override the ecosystem defaults when needed:
|
|
# # preflight = [["go", "version"]]
|
|
# # test_glob = "*_test.go"
|
|
# protect = ["_templ.go"]
|
|
# gate = [
|
|
# ["templ", "generate", "./components/"],
|
|
# ["go", "vet", "./..."],
|
|
# ["go", "build", "./..."],
|
|
# ["golangci-lint", "run"],
|
|
# ["go", "test", "./...", "-count=1"],
|
|
# ]
|
|
# # allow_concurrent = true # let the daemon run multiple tasks for this repo at once instead of
|
|
# # FIFO-serializing them; only safe when concurrent worktrees won't collide
|
|
# # (e.g. different base branches or packages per task). Still bounded by
|
|
# # [containers].max_concurrency globally.
|
|
# # persistent_storage = true # reuse a per-repo Docker volume and Git object cache until volume_ttl
|