# Y Summary Codex > Summarize Codex session JSONL logs into y-schema SummaryRoot outputs with evidence mapping and validation, then attach the summary as a git note under y-codex-session-summary. Use for building or updating the Codex session summarization pipeline or when asked to transform a Codex session log into structured summary artifacts. Use when this capability is needed. - Skill: `tomevault-io/y-summary-codex` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/y-summary-codex` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/y-summary-codex/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/y-summary-codex --- # y-summary-codex ## Purpose Build the summarization pipeline that reads Codex session `.jsonl` logs and emits validated `SummaryRoot` JSON, then attaches the final summary as a git note in namespace `y-codex-session-summary`. ## Inputs - Session log path (absolute or relative to `~/.codex/sessions`). - Output directory for `summary.json`. ## Non-negotiable Guardrails - **YOU must NOT summarize your own progress creating the summary of the work you are reviewing.** - **YOU must only summarize the work performed in the session being reviewed.** ## Required Workflow Follow these steps in order and keep outputs deterministic. 1. Parse session JSONL - Read line-delimited JSON. - Normalize core event model: timestamp, type, actor/source, payload, call id, line index. - Preserve order using `(timestamp, line_index)`. 2. Build evidence map first - Create deterministic evidence IDs from `(timestamp, event_type, snippet, line_index)`. - Store evidence tuples with: timestamp, event, snippet, source. - Do not use assistant reasoning as sole evidence for facts. 3. Extract all 11 sections - Implement extractors using the methods in `references/section-methods.md`. - Preserve fact vs inference (set `basis` accordingly). - Every evidence-backed claim must include `evidence_refs`. - If data is missing, emit empty structures (not omitted) with low-confidence claims only when explicitly allowed. 4. Assemble SummaryRoot - Set `schema_version` to `y-schema::CURRENT_SCHEMA_VERSION`. - Populate all sections (never omit required sections). - Link evidence by ID only. 5. Validate before writing outputs - Use `y-schema::parse_summary` or `validate_summary` and fail with actionable errors (JSON path + reason). - Prefer `validate_summary` for multi-error reporting. - Always run the CLI validator `y validate summary` against the final JSON before writing the git note. - If CLI validation fails, stop and surface the error paths; do not write outputs or notes. - Keep library validation too; CLI validation is the required final gate. 6. Write summary.json - Write `summary.json` from `SummaryRoot` only. - Do not emit human-readable outputs unless explicitly requested by the user. 7. Attach git note - Write the final summary JSON or a compact pointer (see `references/output-contract.md`) as a git note under namespace `y-codex-session-summary`. - Use existing git helper in `/Users/ramos/yapp/y/src/git.rs` if applicable. ## Determinism Rules - Sort lists by first evidence timestamp, then stable tie-breaker (line index or path). - Hash-based IDs must be reproducible from content, not random. - Avoid nondeterministic iteration over hash maps. ## When to read extra references - Use `references/section-methods.md` for extraction heuristics and section-specific rules. - Use `references/output-contract.md` for output formatting and inbox summary structure. - Use `/Users/ramos/yapp/y-skills/y-search-codex/SKILL.md` if you need to locate or search sessions. --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/eqtylab) — claim your Tome and manage your conversions.