agents.toml: re-add cf48 (opus cfold review) dropped during the launch-system migration
The unification transcribed phases from .phases-spec before cf48 was added, so the operator's just-requested opus 4.8 cfold review got dropped. Re-append it after ghost (system is past cf55/on pvfix, so can't insert before pvfix without shifting the live phase index). agents.py re-reads config each tick.
This commit is contained in:
autonomic-bot
@ -618,3 +618,30 @@ session cc-ci-orchestrator-stale can be killed; recipe-mirrors org still private
|
||||
when cf48 starts since per-phase backend isn't overridable).
|
||||
- Also this session: relaunched after a session restart (NOT a host reboot, 13d uptime); loops+watchdog
|
||||
were stopped in a claude-sonnet limit window on cf55 → restarted via RESUME_PHASE=1 launch.py start.
|
||||
|
||||
## 2026-06-13 ~05:27 — launch system unified (agents.toml + agents.py); cutover done
|
||||
- Replaced the 5 bespoke launchers + ~15 dotfiles with ONE config (cc-ci-plan/agents.toml)
|
||||
+ ONE driver (cc-ci-plan/agents.py: up/down/status/watchdog/logs/phase). Design + behavior
|
||||
mapping in cc-ci-plan/plan-unified-launch.md.
|
||||
- launch.py and launch-orchestrator.sh are now COMPATIBILITY SHIMS → agents.py (originals at
|
||||
*.orig). So `launch.py start|status|stop|watchdog|logs`, the systemd boot chain
|
||||
(cc-ci-loops-start → launch.sh → launch.py start), and your startup routine all drive the
|
||||
new system transparently — no behavior change for you.
|
||||
- Config is the single source of truth; the watchdog re-reads it every tick (no more env-vs-file
|
||||
drift, which had caused the opencode-revert bug earlier today). Backend/model/prompt/watch
|
||||
policy per agent live in agents.toml. To change a model or backend: edit agents.toml.
|
||||
- State (phase index, resume ids, limit windows) now under .cc-ci-logs/state/. Phase machine
|
||||
unchanged; de-duped the doubled `mailu` entry (cf55 was idx 10 → now idx 9; current phase
|
||||
pvfix = idx 10). All agents respawned via the new system and confirmed working on pvfix.
|
||||
- The new watchdog tmux session (cc-ci-watchdog) runs `agents.py watchdog`. Same heal/limit/
|
||||
stall/handoff/phase-advance/wake behavior, lifted verbatim, now config-driven.
|
||||
|
||||
## 2026-06-13 ~05:30 — startup: unified agents.toml live; re-added dropped cf48
|
||||
- Session relaunch (NOT a host reboot, 13d uptime). Supervision UP: unified `agents.py watchdog`
|
||||
(--config agents.toml) + builder/adv on claude-sonnet + orchestrator on claude-opus-4-8. Phase
|
||||
pvfix [proxy /16 fix] in progress; cf55 confirmed ## DONE (advance was legit).
|
||||
- The launch-system unification (agents.toml + agents.py) was deployed in the gap. It was transcribed
|
||||
from .phases-spec BEFORE I added cf48 (05:15), so cf48 (the operator's opus cfold review) was
|
||||
DROPPED. Re-added it to agents.toml — appended AFTER ghost (the system is already past cf55/on
|
||||
pvfix, so inserting before pvfix would shift the live phase index). agents.py re-reads config every
|
||||
tick, so no watchdog bounce needed. cf48 runs as the last phase, opus 4.8, claude backend.
|
||||
|
||||
149
cc-ci-plan/agents.toml
Normal file
149
cc-ci-plan/agents.toml
Normal file
@ -0,0 +1,149 @@
|
||||
# cc-ci unified agent configuration — the single source of truth.
|
||||
#
|
||||
# One file declares: which agents exist, their backend, model, prompt, kind, and how the
|
||||
# watchdog supervises them. Read by agents.py (driver + watchdog). Runtime state (phase
|
||||
# index, resume ids, limit windows) lives under <log_dir>/state/, NOT here.
|
||||
#
|
||||
# Precedence: this file is authoritative. A one-off CLI override (env AGENT_<field>_<name>)
|
||||
# affects only a single `agents.py` invocation; the persisted watchdog always reads this file.
|
||||
|
||||
# ─────────────────────────── global watchdog cadence ───────────────────────────
|
||||
[watchdog]
|
||||
signal_interval = 30 # s between handoff / stall / limit checks (light)
|
||||
heavy_interval = 300 # s between heal / phase-advance checks
|
||||
limit_probe_fallback = 300 # flat probe cadence when a reset time can't be parsed
|
||||
limit_reset_slack = 45 # s past a parsed reset before probing
|
||||
|
||||
# ─────────────────────────── backends (declared as data) ───────────────────────────
|
||||
[backend.claude]
|
||||
bin = "claude"
|
||||
flags = "--dangerously-skip-permissions"
|
||||
remote_control = true
|
||||
supports_resume = true
|
||||
prompt_delivery = "arg" # full prompt passed as a CLI argument
|
||||
submit_key = "Enter"
|
||||
stall_idle = 300
|
||||
active_re = "esc to interrupt|⠋|⠙|⠹|⠸|⠼|⠴|⠦|⠧|⠇|⠏|Running tool|▣|Build ·|· \\d+"
|
||||
limit_re = "spend limit|usage limit|limit reached|reached your .*limit|out of (credits|tokens)"
|
||||
fatal_re = "redacted_thinking|blocks cannot be modified|cannot be modified"
|
||||
|
||||
[backend.opencode]
|
||||
bin = "/home/loops/.local/bin/opencode"
|
||||
server = "http://127.0.0.1:4096"
|
||||
supports_resume = false
|
||||
prompt_delivery = "ping" # send after the TUI connects
|
||||
connect_delay = 12
|
||||
submit_key = "C-m"
|
||||
preamble = "set -a; . /srv/cc-ci/.testenv; set +a"
|
||||
stall_idle = 900
|
||||
active_re = "esc interrupt|thinking|inferring|running tool|tool call|preparing patch|reading|searching"
|
||||
limit_re = "spend limit|usage limit|limit reached|reached your .*limit|out of (credits|tokens)"
|
||||
fatal_re = "redacted_thinking|blocks cannot be modified|cannot be modified"
|
||||
|
||||
# ─────────────────────────── defaults for every agent ───────────────────────────
|
||||
[defaults]
|
||||
backend = "claude"
|
||||
model = "claude-sonnet-4-6"
|
||||
dir = "/srv/cc-ci-orch"
|
||||
watch = "heal" # none | heal | heal+stall
|
||||
log_dir = "/srv/cc-ci/.cc-ci-logs"
|
||||
|
||||
# ─────────────────────────── agents ───────────────────────────
|
||||
|
||||
[[agent]]
|
||||
name = "orchestrator" # tmux session: cc-ci-orchestrator
|
||||
kind = "persistent"
|
||||
model = "claude-opus-4-8"
|
||||
resume = true # claude --resume <state/orchestrator.id>
|
||||
watch = "heal" # restart if dead / FATAL / backend-mismatch; never stall-reboot
|
||||
wake = { interval = 3600, prompt_file = "ai-progress-monitor-prompt.txt" }
|
||||
prompt = """
|
||||
STARTUP (auto-launch): you are the cc-ci orchestrator, just (re)launched, likely after a reboot. \
|
||||
Do your AGENTS.md On-startup routine NOW: read cc-ci-plan/REBOOTS.md and run cc-ci-plan/launch.py \
|
||||
status, then send a proactive PushNotification that you are online with the current phase and \
|
||||
reboot count, and confirm cc-ci-loops.service brought the loops + watchdog back (relaunch with \
|
||||
RESUME_PHASE=1 cc-ci-plan/launch.py start if not). Also read cc-ci-plan/JOURNAL.md for recent \
|
||||
context before resuming supervision."""
|
||||
|
||||
[[agent]]
|
||||
name = "builder" # tmux session: cc-ci-builder
|
||||
kind = "loop"
|
||||
role = "builder" # phase kickoff = phase preamble + prompts/builder.md
|
||||
dir = "/srv/cc-ci/cc-ci"
|
||||
watch = "heal+stall"
|
||||
|
||||
[[agent]]
|
||||
name = "adversary"
|
||||
session = "cc-ci-adv" # established convention is abbreviated (logs, remote-control)
|
||||
kind = "loop"
|
||||
role = "adversary"
|
||||
dir = "/srv/cc-ci/cc-ci-adv"
|
||||
watch = "heal+stall"
|
||||
|
||||
[[agent]]
|
||||
name = "assistant" # tmux session: cc-ci-assistant
|
||||
kind = "persistent"
|
||||
resume = true
|
||||
watch = "none"
|
||||
enabled = false
|
||||
prompt = """
|
||||
You are the cc-ci ASSISTANT. Read cc-ci-plan/JOURNAL.md for context, then wait for a specific \
|
||||
plan/task from the orchestrator or operator. Work autonomously until the assigned task is \
|
||||
complete, report the result, and then wait for the next assignment."""
|
||||
|
||||
[[agent]]
|
||||
name = "upgrader" # tmux session: cc-ci-upgrader
|
||||
kind = "task" # one-shot: run to completion, then idle
|
||||
dir = "/srv/cc-ci"
|
||||
watch = "none"
|
||||
enabled = false # launched on demand or by [loop].on_complete
|
||||
prompt = """
|
||||
You are the cc-ci weekly UPGRADER. Invoke the /upgrade-all skill in DEFAULT mode (read \
|
||||
/srv/cc-ci/.claude/skills/upgrade-all/SKILL.md for the full procedure), run it to completion, \
|
||||
then report the summary of every PR opened and go idle."""
|
||||
|
||||
[[agent]]
|
||||
name = "report" # tmux session: cc-ci-report
|
||||
kind = "task"
|
||||
dir = "/srv/cc-ci"
|
||||
model = "claude-opus-4-8" # report is authored by opus even when the upgrader runs sonnet
|
||||
watch = "none"
|
||||
enabled = false
|
||||
prompt = """
|
||||
You generate the public weekly "Recipe Report". Run the /recipe-report skill now (full spec: \
|
||||
/srv/cc-ci/.claude/skills/recipe-report/SKILL.md; creds in /srv/cc-ci/.testenv), then go idle."""
|
||||
|
||||
# Non-AI helper services (started by `up`, not AI sessions)
|
||||
[[service]]
|
||||
name = "cleanlogs" # tmux session: cc-ci-cleanlogs
|
||||
command = "python3 cc-ci-plan/agent-log.py follow-all"
|
||||
dir = "/srv/cc-ci-orch"
|
||||
|
||||
# ─────────────────────────── the phase machine (kind="loop" agents) ───────────────────────────
|
||||
[loop]
|
||||
state_file = "phase-idx" # under <log_dir>/state/
|
||||
resume_phase = true # keep current index across restarts (don't reset to 0)
|
||||
auto_advance = true
|
||||
done_marker = "## DONE"
|
||||
handoff = { repo = "/srv/cc-ci/cc-ci", claim_pings = "adversary", review_pings = "builder", inboxes = ["ADVERSARY-INBOX.md", "BUILDER-INBOX.md"] }
|
||||
on_complete = { trigger_file = ".run-upgrade-on-complete", run = "upgrader" }
|
||||
|
||||
# Transcribed verbatim from the live .phases-spec, with the duplicate `mailu` (old idx 7)
|
||||
# REMOVED. Per-phase model overrides (were .loop-model[-adv]-<id> files) are inline.
|
||||
# Current phase after de-dupe: cf55 (index 9) — see state/phase-idx.
|
||||
phases = [
|
||||
{ id = "rcust", plan = "recipe-custom-restructure-full-plan.md", status = "STATUS-rcust.md" },
|
||||
{ id = "shot", plan = "plan-phase-shot-screenshots.md", status = "STATUS-shot.md" },
|
||||
{ id = "lvl5", plan = "plan-phase-lvl5-lint-rung.md", status = "STATUS-lvl5.md" },
|
||||
{ id = "bsky", plan = "plan-phase-bsky-fix.md", status = "STATUS-bsky.md" },
|
||||
{ id = "dstamp", plan = "plan-phase-dstamp-discourse-drift.md", status = "STATUS-dstamp.md", models = { builder = "opus" } },
|
||||
{ id = "mailu", plan = "plan-phase-mailu-backup.md", status = "STATUS-mailu.md" },
|
||||
{ id = "kuma", plan = "plan-phase-kuma-monitor.md", status = "STATUS-kuma.md" },
|
||||
{ id = "drone", plan = "plan-phase-drone-enroll.md", status = "STATUS-drone.md" },
|
||||
{ id = "cfold", plan = "plan-phase-cfold-custom-folder.md", status = "STATUS-cfold.md" },
|
||||
{ id = "cf55", plan = "plan-phase-cf55-gpt55-cfold-review.md", status = "STATUS-cf55.md", models = { builder = "claude-sonnet-4-6", adversary = "claude-sonnet-4-6" } },
|
||||
{ id = "pvfix", plan = "plan-phase-pvfix-swarm-proxy.md", status = "STATUS-pvfix.md" },
|
||||
{ id = "pvcheck", plan = "plan-phase-pvcheck-post-proxy-verification.md", status = "STATUS-pvcheck.md" },
|
||||
{ id = "ghost", plan = "plan-phase-ghost-reeval.md", status = "STATUS-ghost.md" },
|
||||
{ id = "cf48", plan = "plan-phase-cf48-opus-cfold-review.md", status = "STATUS-cf48.md", models = { builder = "claude-opus-4-8", adversary = "claude-opus-4-8" } },
|
||||
]
|
||||
Reference in New Issue
Block a user