zj-loop-cost

Estimate daily token spend for agentic loop working patterns by cadence and readiness level (L1–L3).

Uses cost metadata from patterns/registry.yaml. When run with a project root, the project-local registry is preferred over the packaged default.

Install & Run

npx @jununfly/zj-loop-cost . --pattern ci-sweeper --cadence 15m --level L2
npx @jununfly/zj-loop-cost . --pattern daily-triage --level L1 --json
npx @jununfly/zj-loop-cost --registry patterns/registry.yaml --pattern daily-triage
npx @jununfly/zj-loop-cost --package-registry --pattern daily-triage
npx @jununfly/zj-loop-cost --list

If npx stalls on package resolution, use an installed binary or npm's local cache:

zj-loop-cost . --pattern daily-triage --level L1
npm exec --offline --package=@jununfly/zj-loop-cost -- zj-loop-cost . --pattern daily-triage --level L1

From this repo:

cd tools/zj-loop-cost
npm install
npm test

Options

| Flag | Description | |------|-------------| | --pattern | Pattern id (see --list) | | --registry | Explicit registry file | | --package-registry | Use packaged registry defaults first | | --cadence | Override cadence (e.g. 15m, 1d) | | --level | L1, L2, or L3 cost/readiness model (default L1) | | --conservative | Use slower cadence from ranges | | --json | Machine-readable output |

Scenarios

Each estimate includes:

• Early-exit / no-op — empty watchlist, minimal tokens
• Full triage — every run does a full scan
• Action every run — implementer + verifier every time (worst case)
• Realistic blend — level-based mix (documented in output)

Pair with zj-loop/zj-loop-budget.md (scaffolded by zj-loop-init) and zj-loop-audit cost observability checks.

Cost level is not execution authority. Route Table policy and pattern-specific guards decide whether a loop is report-only, request-only, dry-run, or live.

See docs/operating-loops.md.