reporting-trace-maintainer

Maintain report-ready work traces while Codex executes tasks, research, repo changes, experiments, or project work. Use when the user asks to keep work logs, prepare future advisor/stakeholder reports

SKILL.md

GitHub

Reporting Trace Maintainer

Maintain a light trace layer that can later become an advisor update, stakeholder briefing, weekly report, or project status note. The trace is not the final report; it is the structured raw material behind one.

Core Idea

During task execution, preserve the chain:

last plan -> work done -> evidence -> interpretation -> next plan -> questions

This mirrors effective research/advisor update practice: make progress visible, expose blockers early, and leave clear next actions without turning every task into a polished report.

Workflow

  1. Locate the trace convention.

    • In a repo, first look for reports/work-traces/README.md, codex/workflows/reporting-trace-contract.md, project INDEX.md, task graph files, and codex/project-map.md.
    • If no convention exists and the user asked for durable traces, create or suggest a minimal reports/work-traces/ structure.
  2. Decide whether a trace is needed.

    • Write or update a trace when the task changes project state, produces evidence, resolves a blocker, creates a decision point, or will likely need advisor/stakeholder reporting.
    • Skip durable trace updates for trivial commands, private credentials, purely local cleanup, or work that the user explicitly wants to keep out of long-term notes.
  3. Choose the trace location.

    • Prefer reports/work-traces/<project>/<YYYY-MM-DD>-<task-id-or-slug>.md.
    • If a project already has a local trace convention, follow it.
    • Keep bulky artifacts in the project/artifact repo and link them; do not copy large logs, PDFs, screenshots, datasets, or raw private material into the trace.
  4. Record only report-useful facts.

    • Include previous plan, work performed, evidence links, interpretation, decisions or decision needs, blockers, next plan, and questions.
    • Separate facts, inference, user preference, and pending verification.
    • Use repo-relative paths when possible; use external paths only when the repo already permits them or the path is the execution boundary.
  5. Keep source-of-truth files aligned.

    • If task state changed, update the task graph or project index according to the repo contract.
    • If a decision was made, route to decision-record-writer or add a decision pointer.
    • If evidence was collected, update the relevant evidence manifest or project evidence file.
  6. Close each task with a trace delta.

    • Mention the trace path in the final response when you created or updated one.
    • Do not claim a final report exists unless you actually wrote one.

Trace Sections

Use this compact shape by default:

# Work Trace: <Task Title>

Updated: YYYY-MM-DD
Project: <project-id>
Task: <task-id or short slug>
Audience Future Use: advisor-update | stakeholder-status | weekly-review | project-briefing | internal
Status: active | paused | completed | blocked

## Last Plan

- ...

## Work Done

- ...

## Evidence

| Evidence | Path / Source | What It Supports | Confidence |
|---|---|---|---|

## Interpretation

- Fact:
- Inference:
- Uncertainty:

## Decisions / Open Decisions

- ...

## Next Plan

- [ ] ...

## Questions For Next Meeting

- ...

For detailed schema rules and examples, read references/trace-schema.md.

Boundaries

  • Do not store secrets, tokens, cookies, private raw transcripts, or sensitive personal details.
  • Do not turn a trace into a polished outward-facing narrative unless the user asks.
  • Do not invent work, evidence, or decisions to make a trace look complete.
  • Do not duplicate bulky artifacts; link to their maintained location.
  • Do not overwrite user-written project truth. Update narrowly and preserve existing wording when possible.

Handoffs

  • Use project-briefing-room when the user wants a stakeholder-facing project status briefing.
  • Use evidence-synthesis-docs when many traces or evidence records need synthesis.
  • Use decision-record-writer when a tradeoff or workflow choice should become durable.
  • Use codex-deep-interview when the report audience, non-goals, or decision boundary is unclear.