Wasted Cycles — local wall-clock profiler for machine-blocking stalls in AI coding agent loops

TL;DR in plain English

• What it is: Wasted Cycles is a local-first wall-clock profiler that finds where machine time blocks agent loops (builds, tests, CI, containers, packages, sub-agents, and repeated machine work). See the project page: https://zozo123.github.io/wasted-cycles/

• How to try fast: verify the installer checksum, run the installer, and run a short audit. Example installer command: curl -fsSL https://zozo123.github.io/cycles | sh (verify checksum first). A quick profile: wasted-cycles --days 7 (the command saves a JSON export). Source: https://zozo123.github.io/wasted-cycles/

• What you get: wall-clock totals and a short ranked list (example demo totals: AGENT LOOP 4h 59m; BLOCKED 2h 05m; Model work 1h 28m; Build 1h 06m; Tests 47m). Use the top blocker to prioritize work. Source: https://zozo123.github.io/wasted-cycles/

Quick checklist

• [ ] verify installer checksum and store it
• [ ] run wasted-cycles --days 7 and save the JSON export
• [ ] open BREAKDOWN and note the top blocked category
• [ ] create one remediation ticket with baseline numbers

One-line methodology: Wasted Cycles measures elapsed (wall-clock) time and reports machine-blocked time separately from human wait. Source: https://zozo123.github.io/wasted-cycles/

What changed

• Distribution and trust model: the tool is distributed as a checksum-verified binary, runs from a temporary directory, supports version pinning, requires no account, and does not upload traces by default (installer: curl -fsSL https://zozo123.github.io/cycles | sh). Source: https://zozo123.github.io/wasted-cycles/

• Measurement focus: it measures wall-clock elapsed time and classifies blocking sources—builds, tests, CI, containers, packages, sub-agents, repeated machine work—while reporting human wait separately. Source: https://zozo123.github.io/wasted-cycles/

• GitHub Actions support: wasted-cycles github owner/repo --days 30 inspects completed workflow runs and reports queue delay, median and p95 latency, unsuccessful time, and per-workflow elapsed totals. Demo values include WORKFLOW LATENCY 22h 29m; QUEUE DELAY 0s; MEDIAN 38s; P95 33m; SUCCESS RATE 100% (84/84). Source: https://zozo123.github.io/wasted-cycles/

• Trace reading and classification: the tool reads existing traces in place (examples: Codex, Claude Code, Cursor, Grok) and classifies using structured event fields and executed commands; pasted logs/quoted commands are excluded. Source: https://zozo123.github.io/wasted-cycles/

Why this matters (for real teams)

• Focus fixes on real machine stalls. Teams often optimize prompts or orchestration while the real delay is a build or test that blocks the agent loop; Wasted Cycles surfaces elapsed blocking time so you can pick the highest-impact change. Source: https://zozo123.github.io/wasted-cycles/

• Prioritize using measured thresholds. Example early targets: reclaim >=10% of AGENT LOOP elapsed time, reduce BLOCKED totals by >=30%, or cut workflow p95 by 30% (demo p95 = 33m). Use 7-day audits for quick feedback and 30-day windows for CI trends. Source: https://zozo123.github.io/wasted-cycles/

• Comparable baselines across sessions. Sessions are defined so metrics are stable: gaps >2 hours start a new session; shorter gaps cap at 30 minutes and remain visible in the JSON export. Source: https://zozo123.github.io/wasted-cycles/

Concrete example: what this looks like in practice

1. Run: wasted-cycles --days 30 and open BREAKDOWN. Source: https://zozo123.github.io/wasted-cycles/

2. Read totals and pick the top blocker. The scaled demo reports: AGENT LOOP 4h 59m; BLOCKED 2h 05m; Model work 1h 28m; Build 1h 06m; Tests 47m. Source: https://zozo123.github.io/wasted-cycles/

3. One measurable change and re-measure. Example targets you can aim for:

• Reclaim >=10% of AGENT LOOP elapsed time (baseline check).
• Reduce BLOCKED total by >=30% after one change.
• Cut workflow p95 by 30% (demo p95 = 33m) as an initial objective.

Immediate low-effort fixes to try (use the JSON export for exact numbers): enable dependency caching, pin build images to avoid repeated pulls, split long tests into 2–4 parallel shards, or move non-critical scans off-peak. Source: https://zozo123.github.io/wasted-cycles/

What small teams and solo founders should do now

A short, concrete sequence tailored for solo founders and small teams (time estimates and actionable steps). Source: https://zozo123.github.io/wasted-cycles/

1. Fast audit (30–60 minutes)

• Verify the installer checksum, run the installer in a temporary directory, then run: wasted-cycles --days 7. Save the JSON export and note the top 1–2 blocked categories (e.g., Build 1h 06m, Tests 47m). Record AGENT LOOP and BLOCKED totals. Source: https://zozo123.github.io/wasted-cycles/

2. One-ticket, one-change (15–120 minutes)

• Create a single remediation ticket that records baseline BLOCKED and p95. Choose a single low-risk fix that you can implement in one deploy (examples: enable dependency cache, pin build image, add 2–4 test shards). After rollout, re-run the same --days profile and compare BLOCKED and p95. Aim to reclaim >=10% AGENT LOOP or cut BLOCKED by >=30%. Source: https://zozo123.github.io/wasted-cycles/

3. Lightweight repeatable cadence (10–30 minutes per week)

• Automate a weekly wasted-cycles --days 7 export or run it manually, store the JSON alongside CI artifacts, and keep a simple changelog entry: who ran it, date, top blocker, and result. Use monthly 30-day GitHub profiling for CI: wasted-cycles github owner/repo --days 30 to track QUEUE DELAY, MEDIAN and P95. Demo: QUEUE DELAY 0s; MEDIAN 38s; P95 33m. Source: https://zozo123.github.io/wasted-cycles/

Extra lean tips for solos

• If you have one machine: run locally and keep two exports (before/after) to avoid context switching.

• If you have a single CI pipeline: focus on the single workflow with the highest elapsed time (use the per-workflow totals in the JSON).

• Use the one-ticket rule to avoid scope creep: one ticket, one measurable rollback condition.

• [ ] Fast audit done (verify checksum + 7-day run)

• [ ] One remediation ticket created with baseline numbers

• [ ] Weekly export scheduled or logged

Regional lens (UK)

• Local-first control: the binary runs locally, supports checksum verification and version pinning, and does not upload traces by default — useful for UK teams that prefer to keep traces on-prem or in a controlled vault. Source: https://zozo123.github.io/wasted-cycles/

• Measure real queue behavior: run wasted-cycles github owner/repo --days 30 to expose QUEUE DELAY and p95 for your repo (demo QUEUE DELAY 0s; MEDIAN 38s; P95 33m). Store JSON outputs in an internal vault or CI artifact store for auditability. Source: https://zozo123.github.io/wasted-cycles/

UK operational checklist

• [ ] verify installer checksum before running
• [ ] run in an isolated temporary directory
• [ ] store JSON exports in an internal vault or CI artifact store
• [ ] log who ran the audit and when

US, UK, FR comparison

| Jurisdiction | Typical priority | Why Wasted Cycles helps | First recommended step | |---|---:|---|---| | US | cost & dev velocity | Profile Actions p95 and runner use to target billed minutes and speed | wasted-cycles github owner/repo --days 30 | | UK | privacy & operational control | Local-first, checksum-verified binary avoids uploads | verify checksum & run local traces | | FR / EU | compliance & data locality | No-upload design and local traces reduce transfer concerns | run locally and consult legal before sharing reports |

Notes: operational guidance only. Source: https://zozo123.github.io/wasted-cycles/

Technical notes + this-week checklist

Short methodology note: Wasted Cycles measures wall-clock elapsed time; overlapping runs count separately and workflow latency is created_at → updated_at. Source: https://zozo123.github.io/wasted-cycles/

Assumptions / Hypotheses

• The tool reports wall-clock elapsed time and separates human wait from machine-blocked time. Source: https://zozo123.github.io/wasted-cycles/
• Sessions: gaps >2 hours start a new session; shorter gaps are capped at 30 minutes in the JSON. Source: https://zozo123.github.io/wasted-cycles/
• Classification uses structured event fields and executed commands; pasted logs and quoted commands are excluded. Source: https://zozo123.github.io/wasted-cycles/

Risks / Mitigations

• Risk: running the installer without checksum verification. Mitigation: always verify checksum and store it in your artifact registry before running. Source: https://zozo123.github.io/wasted-cycles/
• Risk: conflating elapsed wall-clock time with billing minutes. Mitigation: use elapsed time to prioritize blockers and consult billing reports separately for $ decisions. Source: https://zozo123.github.io/wasted-cycles/
• Risk: over-optimizing low-impact items. Mitigation: follow the one-ticket rule and use thresholds (>=10% AGENT LOOP reclaim, >=30% BLOCKED reduction, 30% p95 cut) as gates.

Next steps

This-week checklist:

• [ ] Verify binary checksum and run a 7-day local profile: wasted-cycles --days 7 (installer: curl -fsSL https://zozo123.github.io/cycles | sh). Save the JSON export. Source: https://zozo123.github.io/wasted-cycles/
• [ ] If you use GitHub Actions, run: wasted-cycles github owner/repo --days 30 and record QUEUE DELAY, MEDIAN, P95, and per-workflow elapsed totals (demo: WORKFLOW LATENCY 22h 29m; MEDIAN / P95 38s / 33m; SUCCESS RATE 100% for 84/84 runs). Source: https://zozo123.github.io/wasted-cycles/
• [ ] Create one remediation ticket for the top blocker with baseline numbers and an acceptance target (e.g., BLOCKED down by >=30% or p95 down by 30%).
• [ ] Implement the change behind a rollout gate and re-run the profiler using the same --days window.
• [ ] Compare JSON exports (before/after) and confirm thresholds before wider rollout.

If you want, I can convert the checklist into a one-page ticket template that records baseline numbers, acceptance criteria, and rollback conditions, or produce shell snippets to verify checksums and archive JSON exports.

Wasted Cycles — local wall-clock profiler for machine-blocking stalls in AI coding agent loops

Local-first wall-clock profiler that identifies machine-blocking stalls in AI agent loops: builds, tests, CI, containers. Run a checksum-verified binary to aud…

https://aisignals.dev/posts/2026-08-20-wasted-cycles-local-wall-clock-profiler-for-machine-blocking-stalls-in-ai-coding-agent-loops

(Weekly: AI news, agent patterns, tutorials)