MOPAC/harness.toml.example

# MOPAC harness configuration, v0 (2026-08-28).
#
# Copy to harness.toml and adjust. harness.toml is gitignored: secrets never
# live in this file, only refs (env:NAME | file:PATH | literal:VALUE |
# mpk:PLACEHOLDER via [keyproxy]; the bw: bitwarden ref lands with the key
# wrapper in build phase 3).

# Vertical / stack identity for this harness instance.
vertical = "demo"

# Where the bash tool executes and REPORTs land (relative to the CWD the
# harness is started from).
work_root = "."
report_dir = "reports"

[loop]
# Bounded turn: max LLM round trips per task (tool calls included).
max_rounds = 8
# `harness loop` daemon: Redmine scan interval and state location
# (append-only loop.jsonl, dedup by issue id + updated_on).
poll_interval_secs = 120
state_dir = "state/loop"

[redmine]
url = "https://rm.example.org"
key_ref = "env:HARNESS_REDMINE_KEY"
# Released scope: raw /issues.json filter params...
scope_query = "project=mopac&status_id=released&limit=25"
# ...or a saved custom query id (scope_query wins when both are set).
# scope_query_id = 42
# Custom field carrying the TASK class; issues without it get default_class.
class_field = "Class"
default_class = "primary"

# `harness loop` status transitions: after a REPORT is noted back on the
# issue, an issue whose CURRENT status matches a key here is moved to the
# value. Statuses are names (resolved via /issue_statuses.json); an empty
# map leaves status alone.
[redmine.status_map]
"In Progress" = "Done"

[litellm]
base_url = "http://192.168.3.78:4001"
key_ref = "env:HARNESS_LITELLM_KEY"
timeout_secs = 120
max_retries = 2

# KEYPROXY (optional): resolve `mpk:` key refs through the ukrrs/mopac-keyproxy
# hop (POST /v1/resolve, bearer auth). Without this section, env:/file:/
# literal: refs keep working as-is — the loop runs with or without keyproxy up.
# [keyproxy]
# url = "http://127.0.0.1:8082"
# token_ref = "env:HARNESS_KEYPROXY_TOKEN"   # local ref only (no mpk: recursion)
# cache_ttl_secs = 60

# GITEA (optional, off by default): `harness loop` commits each REPORT file
# to this repo right after writing it (contents API, create-or-update).
# [gitea]
# url = "https://git.example.org"
# key_ref = "env:HARNESS_GITEA_KEY"
# owner = "ukrrs"
# repo = "MOPAC-reports"
# branch = "main"          # empty = repo default branch
# commit_reports = false

# MODEL ROUTING v0 (static, config-only, no heuristics): [models] is the
# tier map (tier alias -> concrete proxy model); [models.classes] maps task
# classes to tiers. Requests go out with the CONCRETE model name resolved
# here. Swapping models (e.g. glm-4.7-flash -> glm-5.3-flash once it is
# configured on the proxy) is a one-line edit in this file.
[models]
mopac-study   = "glm-4.7-flash"  # flash tier
mopac-code    = "glm-5.2"        # flagship
mopac-review  = "glm-5-turbo"    # mid
mopac-primary = "glm-5.3"        # default / flagship+
mopac-vision  = "glm-4.6v"       # vision when needed
default_tier  = "mopac-primary"

[models.classes]
study        = "mopac-study"
read         = "mopac-study"
code         = "mopac-code"
architecture = "mopac-code"
review       = "mopac-review"
summarize    = "mopac-review"
writeback    = "mopac-review"
vision       = "mopac-vision"
primary      = "mopac-primary"

# Exec tool: allow-listed bash. Org preset: deny-first, read-leaning allow
# list, no sudo/ssh/network exfil. Compound commands are checked segment by
# segment; command substitution and subshells are always denied.
[tools.bash]
enabled = true
timeout_secs = 60
max_output_bytes = 100000
default = "deny"
deny = [
  "sudo *",
  "ssh *",
  "scp *",
  "nc *",
  "curl *",
  "wget *",
  "rm -rf *",
]
allow = [
  "pwd",
  "ls *",
  "cat *",
  "head *",
  "tail *",
  "grep *",
  "find *",
  "wc *",
  "echo *",
  "env",
  "git status",
  "git diff *",
  "git log *",
  "git show *",
  "go version",
  "go build *",
  "go vet *",
  "go test *",
]

# MVP demo (Charles, 2026-08-28 ~19:00): `harness once --demo` runs this
# issue through LiteLLM; the GLM self-description lands as the REPORT.
[demo]
id = "demo-1"
subject = "MVP demo: GLM self-description"
prompt = "tell me about yourself"
class = "primary"

# EVENTS: the `harness events` webhook receiver (Redmine/Discourse/Gitea
# punch the harness; DESIGN "Events are V1 scope"). Secrets are refs only,
# resolved at startup, never logged. At least one secret_ref must resolve
# or the receiver refuses to start.
[events]
listen = ":4100"            # bind address; publish on the LAN via docker -p
state_dir = "state/events"  # append-only events.jsonl + dedup index

[events.redmine]
secret_ref = "env:HARNESS_REDMINE_WEBHOOK_SECRET"
# secret_header = "X-Redmine-Webhook-Secret"   # default; match your plugin

[events.discourse]
secret_ref = "env:HARNESS_DISCOURSE_WEBHOOK_SECRET"
# secret_header = "X-Discourse-Webhook-Secret"  # default

[events.gitea]
secret_ref = "env:HARNESS_GITEA_WEBHOOK_SECRET"
# Gitea always verifies via HMAC-SHA256 in X-Gitea-Signature; the
# secret_header override does not apply to it.

# SERVE (optional): the `harness serve` OpenAI-compatible front door for
# OpenWebUI (DESIGN "OWUI front door"). Each [models.classes] class is
# exposed as a servable model named mopac-<class> (mopac-study,
# mopac-code, ...); POST /v1/chat/completions runs ONE bounded stateless
# conductor turn over the conversation history OWUI sends (tools off,
# v0; non-streaming). Bearer vkey auth — the same value goes into the
# OWUI connection config. Own port: coexists with [events].
[serve]
listen = ":8090"                        # publish on the LAN via docker -p
vkey_ref = "env:HARNESS_SERVE_VKEY"
# enabled_models = ["mopac-primary", "mopac-study"]  # optional subset

# QUOTA GATE (Redmine 490, off by default until verified live): z.ai coding
# plan credit buckets (5h + weekly), back-pressure thresholds, TZ-aware peak
# window, and the redis shared-state hop so all harness instances of one
# account coordinate. The loop DEFERS gated work with a logged reason; it
# never hard-fails. With no usage_url it runs on locally estimated
# consumption against the configured plan limits (see README: z.ai has no
# public usage endpoint yet — LIVE VERIFICATION open).
# [quota]
# enabled = true
# account = "zai-max-1"           # plan label shared across instances
# plan_5h_credits = 28000         # Max plan; Lite 2000/10000, Pro 12000/60000
# plan_weekly_credits = 140000
# usage_url = ""                  # set when z.ai ships the endpoint
# key_ref = "env:HARNESS_ZAI_KEY" # bearer for usage_url; never logged
# poll_interval_secs = 300
# defer_at_pct = 85               # heavy classes defer, flash tier continues
# block_at_pct = 95               # everything defers until reset (the wall)
# # Peak = z.ai Mon-Fri 14:00-18:00 Singapore == 01:00-05:00 America/Chicago
# # in winter (00:00-04:00 during US DST — adjust in March/November).
# peak_start = "01:00"
# peak_end = "05:00"
# timezone = "America/Chicago"
# peak_weekdays_only = true
# peak_classes = ["study", "read"]  # flash/LLM-lite classes allowed in peak
# # Shared state for the 9 instances across 2 hosts (redis docker container
# # on 192.168.3.78; see README runbook). Empty = local-only estimates.
# redis_url = "redis://192.168.3.78:6390/0"

# RESOURCE GATE (Redmine 491, off by default): read-only host monitor —
# loadavg, mem available, work_root disk free, IO pressure (/proc/pressure
# io, skipped when PSI is absent). The loop defers dispatch while busy.
# cgroup enforcement is deploy-time (README runbook).
# [resources]
# enabled = true
# max_load_avg = 6.0
# min_mem_available_mb = 2048
# min_disk_free_mb = 5120
# max_io_delay_pct = 90.0