static bake — every format quoted from its source file on 2026-07-09

Logging discipline

Four places this fleet writes logs, each with a machine-parseable format that was designed before the consumer that reads it. Everything below is quoted from the named source file — none of it is an idealized example.

1 · deploy-runner step markers

From plugins/cwc-deploy/skills/deploy-runner/deploy_runner/script.py (render_script()). Every rendered deploy script appends these lines per step; the hard rule (from log.py's docstring) is that success is read from an explicit RUN_EXIT=0 line, never inferred from session liveness — a log with clean-looking steps and no RUN_EXIT line means RUNNING, not done.

STEP={i} NAME={step.name} START ts=$(date +%s)
STEP={i} NAME={step.name} EXIT=$rc DURATION_S=${_step_{i}_dur} ts=$(date +%s)
RUN_EXIT=$rc ts=$(date +%s)   # emitted early, on the first failing step
RUN_EXIT=0 ts=$(date +%s)     # terminal line of a fully clean run

And the reader side — the exact regexes deploy_runner/log.py parses those markers with (writer and parser live in the same package, tested together):

_STEP_EXIT_RE = r"^STEP=(?P<step>\d+) NAME=(?P<name>\S+) EXIT=(?P<exit>-?\d+) DURATION_S=(?P<dur>[\d.]+)"
_RUN_EXIT_RE  = r"^RUN_EXIT=(?P<exit>-?\d+)"

2 · worker-build.sh — the kit drift gate

From scripts/worker-build.sh. Exit codes: 0 = build OK (a missing lock is warned, not failed), 1 = tsc failure, 2 = drift gate FAILED. The literal output lines:

literal output	meaning
`OK: typecheck clean`	npx tsc --noEmit passed for the worker
`FAIL: typecheck failed for workers/$worker`	tsc failed — exit 1
`FAIL: kit drift — these kit files changed since $worker's last recorded build:`	a kit file this worker imports changed after its last recorded build — exit 2, drifted files listed one per line
`OK: drift gate clean — kit unchanged since <locked_at>`	every kit dep's hash matches the worker's .build-lock.json
`WARN: $worker has no .build-lock.json — no lock — run with --record after a verified deploy`	warned, not failed — exit 0
`OK: no stale workers — every locked worker matches current kit hashes`	fleet mode (--stale) clean

3 · crawl_log — the crawl audit table

From crates/durable-store/migrations/postgres/001_schema.sql (lines 39–47). Every ATS crawl attempt — changed or not — gets a row: the sha256 of the board snapshot, whether it changed, how many jobs, and how long the crawl took.

CREATE TABLE IF NOT EXISTS crawl_log (
    id              BIGSERIAL   PRIMARY KEY,
    board_token     TEXT        NOT NULL,
    crawled_at      TIMESTAMPTZ DEFAULT NOW(),
    snapshot_sha256 TEXT        NOT NULL,
    changed         BOOLEAN     DEFAULT FALSE,
    job_count       INTEGER     DEFAULT 0,
    duration_ms     INTEGER     DEFAULT 0
);
-- 002_indexes.sql:
CREATE INDEX IF NOT EXISTS idx_crawl_board ON crawl_log(board_token, crawled_at DESC);

4 · [observability] in wrangler.toml

Cloudflare Workers observability (structured logs in the dashboard) is on for 88 of 97 workers/*/wrangler.toml files (grep -l '\[observability\]' workers/*/wrangler.toml | wc -l, re-run 2026-07-10 — a repo-only fact, dated snapshot; the totals grow as new workers are scaffolded, since every scaffolded wrangler.toml ships the block). The stable part of that count is the 9 files without it — unchanged since the 2026-07-09 count, all predating the shared scaffold:

agentcrawls cron etl subagentcowork subagentcoworkers subagentdb subagentetl subagentworkers web

Honest scope