ワンクリックでManusで任意のスキルを実行

始める

codex-project-insights

Cross-session audit using weighted scoring to generate ranked rule integrations.

Manusで実行

概要

Cross-session audit using weighted scoring to generate ranked rule integrations.

インストールコマンド

npx skills add https://github.com/hamanpaul/custom-skills --skill codex-project-insights

このコマンドをClaude Codeにコピー＆ペーストしてスキルをインストール

ソース

hamanpaul/custom-skills

スター0

フォーク0

更新日2026年4月6日 13:27

ファイルエクスプローラー

7 ファイル

SKILL.md

readonly

name	codex-project-insights
description	Cross-session audit using weighted scoring to generate ranked rule integrations.

codex-project-insights

Trigger

Use for periodic project-level audits (manual or cron).

Inputs

~/.codex/sessions/*.jsonl (default --sessions-root)
~/.copilot/session-state/*/ (via --extra-sessions-root)
~/.codex/history.jsonl

Scoring

half_life_days = 21
score = 0.55*short_window + 0.30*long_window + 0.15*severity
short_window = 14 days
long_window = 90 days
add threshold = 0.65 (2 consecutive cycles)
remove threshold = 0.25 (4 consecutive cycles)

Workflow

Aggregate event signals across sessions.
Apply time decay and scoring model.
Rank signals by score.
Generate integration candidates with threshold decisions.
Respect cooldown and max rules per cycle.
Integrate accepted rules directly into .agents/AGENTS.md.
If running full cycle, enforce commit guard before auto-commit.

Tooling

Standalone scoring:
- .agents/skills/custom/codex-project-insights/scripts/project_insights.py
Full cycle orchestration:
- .agents/skills/custom/codex-project-insights/scripts/self_evolve_cycle.py

Output

insights-<timestamp>.json
integration-<timestamp>.json
cycle-<timestamp>.json (if full cycle)
ranked rows with score
add/remove candidate sets and integration results

このリポジトリの他の Skills

同じリポジトリ

project-usage-assess

hamanpaul/custom-skills

Aggregate per-project Copilot premium requests and Codex token usage from local session logs, then generate a markdown table with totals for usage/quota assessment.

2026-04-140

broadcom-prpl-build-code

hamanpaul/custom-skills

Guide for building and independently verifying Broadcom BGW720 images after prplWare patch integration. Use this when checking remote freshness, building in Docker, monitoring logs, and validating artifacts in both the working tree and a fresh verification clone.

2026-04-090

broadcom-prpl-patch

hamanpaul/custom-skills

將 Broadcom prplWare patch bundle 套用到 BGW720 worktree 的流程。用於比較新舊 patch drop、比對目前 repo 差異、判斷哪些檔案可直接覆蓋，以及依 `notes.txt` 或 release note 決定 impl107 helper driver patch 要套進 source 還是只保留為 repo artifact。

2026-04-090

evolve

hamanpaul/custom-skills

Run full self-maintenance cycle (insights + lesson + ACP decision + AGENTS integration + optional commit) across Codex and Copilot sessions.

2026-04-060

smux-manager-skill

hamanpaul/custom-skills

Run a smux-native manager/worker workflow over tmux panes with task envelopes, sanitized context mirrors, and optional thin helper scripts.

2026-04-060

coordinator

hamanpaul/custom-skills

Orchestrate multi-agent short-lived jobs with run/jobs/stat/stop/kill/cron-run plus relay send/replay.

2026-04-040

ソース

hamanpaul

hamanpaul/custom-skills

GitHub リポジトリを開く Creator のリポジトリを見る

インストールコマンド

ダウンロード

Manusで実行

役立つ用途SOC

ソフトウェア開発者コンピュータ・数学職15-1252L4

name	codex-project-insights
description	Cross-session audit using weighted scoring to generate ranked rule integrations.

codex-project-insights

Trigger

Use for periodic project-level audits (manual or cron).

Inputs

~/.codex/sessions/*.jsonl (default --sessions-root)
~/.copilot/session-state/*/ (via --extra-sessions-root)
~/.codex/history.jsonl

Scoring

half_life_days = 21
score = 0.55*short_window + 0.30*long_window + 0.15*severity
short_window = 14 days
long_window = 90 days
add threshold = 0.65 (2 consecutive cycles)
remove threshold = 0.25 (4 consecutive cycles)

Workflow

Aggregate event signals across sessions.
Apply time decay and scoring model.
Rank signals by score.
Generate integration candidates with threshold decisions.
Respect cooldown and max rules per cycle.
Integrate accepted rules directly into .agents/AGENTS.md.
If running full cycle, enforce commit guard before auto-commit.

Tooling

Standalone scoring:
- .agents/skills/custom/codex-project-insights/scripts/project_insights.py
Full cycle orchestration:
- .agents/skills/custom/codex-project-insights/scripts/self_evolve_cycle.py

Output

insights-<timestamp>.json
integration-<timestamp>.json
cycle-<timestamp>.json (if full cycle)
ranked rows with score
add/remove candidate sets and integration results