measure-development-run
Measure a real Codex development task retrospectively from its first user prompt through the prompt that invokes this skill. Use when the user asks to quantify delivery time, tokens, agents, tool failures, rework, scope completion, functional evidence, code or architecture or security quality, intervention, cost, lessons, or improvement opportunities for the current or an archived Codex task.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- measure-development-run
- description
- Measure a real Codex development task retrospectively from its first user prompt through the prompt that invokes this skill. Use when the user asks to quantify delivery time, tokens, agents, tool failures, rework, scope completion, functional evidence, code or architecture or security quality, intervention, cost, lessons, or improvement opportunities for the current or an archived Codex task.
02 资源文件 4
- 参考文档agents/openai.yaml281 B
- 资产文件assets/assessment-template.json1.9 KB
- 参考文档references/measurement-contract.md5.1 KB
- 脚本scripts/measure-development-run.mjs69.5 KB
03 SKILL.md 正文
Measure Development Run
Create one private, append-only evidence snapshot for the current Codex task. Analyze all work types, including planning, diagnosis, implementation, review, QA, release, and mixed runs.
Measure first; interpret second. Never turn missing telemetry into an estimate.
Run the measurement
-
Read
references/measurement-contract.md. -
Resolve the installed skill directory:
measurement_skill_dir="${CODEX_HOME:-$HOME/.codex}/skills/measure-development-run" -
Create an isolated working directory:
measurement_work_dir="$(mktemp -d)" -
Collect deterministic telemetry. The default cutoff is the latest user prompt, which is the invocation boundary:
node "$measurement_skill_dir/scripts/measure-development-run.mjs" collect \ --output "$measurement_work_dir/telemetry.json" -
Create the assessment template:
node "$measurement_skill_dir/scripts/measure-development-run.mjs" template \ --output "$measurement_work_dir/assessment.json" -
Inspect the telemetry, the visible conversation, repository evidence, and any relevant read-only Git, verification CLI, CI, PR, preview, production, or tracker evidence. If compaction hides necessary history, inspect only the required user and assistant records from the exact
source.sessionPathreported by telemetry. Do not copy or persist a full transcript. -
Replace every placeholder in
assessment.json. Use direct evidence. Mark unavailable evidence asunavailable; do not infer success from implementation, labels, or file presence. -
Finalize the immutable JSON and Markdown reports:
node "$measurement_skill_dir/scripts/measure-development-run.mjs" finalize \ --assessment "$measurement_work_dir/assessment.json" -
Validate the emitted JSON path:
node "$measurement_skill_dir/scripts/measure-development-run.mjs" validate \ --measurement "<absolute-json-path>" -
Return measured work plus attributable external wait as the headline duration, followed by agent operational time, external wait, outcome, biggest loss, and saved report paths. Treat thread span only as context. Do not claim the skill is complete until both files exist and validation passes.
Preserve the boundary
- Measure the entire Codex thread/task from its first real user prompt through the user prompt that invoked the skill. Do not reset the boundary after scope changes, resumptions, or additional implementation phases in the same thread.
- Exclude the measurement turn itself. Use
--cutoff nowonly for development or explicit forensic backfills where no invocation prompt exists, and disclose that boundary. - Do not report task age or thread span as delivery duration. Agent operational time is completed plus in-progress parent-turn time and already includes waits for tools and child agents within those turns.
- For every
externalTimingentry, setverifiedexplicitly and setoverlaptoinside-active-turn,outside-active-turn, orunknown. Verified outside-turn entries require non-empty evidence plus exactstartedAtandendedAttimestamps;durationMsmust match that interval. The finalizer rejects overlap with active turns and unions concurrent outside intervals. Keep unknown overlap separate. - Treat between-turn gaps as unattributed idle time and exclude them from the headline unless direct external evidence records a non-overlapping wait.
- Keep controlled model benchmarks separate from real-work measurements.
- Count parallel agent time as agent-minutes, not additional elapsed time.
- Separate product defects, verification defects, infrastructure blockers, permission blockers, harness limitations, scope changes, and human waits.
- Treat outcome correctness and verified quality as prior to scope, speed, and cost.
- Do not create one composite score.
- Never store raw prompts, assistant messages, reasoning, tool inputs, tool outputs, secrets, credentials, or environment values in the measurement artifacts.
- Preserve every prior snapshot. A correction must set
supersedesto an earlier measurement ID and create new files.
Require an honest retrospective
Always finish the Markdown report with concrete recommendations. Include what worked, what failed, time loss, rework, missing evidence, and one to five changes whose expected impact and validation method are explicit.
04 同仓库更多 Skills
working-backwards
Define a software feature through PRFAQ and staged documents when the user requests working backwards before coding.
flow-implement
Implement a requested software change through verification, corrections, and its authorized delivery endpoint.
development-steward
Run the scheduled, read-only weekly maintenance review for the six explicitly opted-in repositories and prepare one concise private report for Check-in.
grill-with-docs
Run one recoverable interview for functional, visual, or mixed software grills; reuse equivalent answers and offer the shared HTML questionnaire.
global-agent-guardrails
Install, audit, test, or update the Development System's global destructive-command guard for Codex, Claude Code and T3 Code. Use when configuring agent safety hooks or investigating whether dangerous shell operations are blocked.
impeccable
Design or refine frontend UI using project context and a focused design playbook; also handles visual critique and polish.