task-observer: new docs/ai/long-running-workstream from 2026-06-15 - #55
task-observer: new docs/ai/long-running-workstream from 2026-06-15#55divad12 wants to merge 1 commit into
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 662e8f2c96
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| | Editing AGENTS.md, .agents/skills/, docs/ai/ | writing-docs.md | | ||
| | Maintaining or troubleshooting ask-intern | ask-intern.md | | ||
| | Operating or changing the learning system | learning-system.md | | ||
| | Long-running multi-session or multi-compaction workstreams | long-running-workstream.md | |
There was a problem hiding this comment.
This registers the new doc for root-session adaptive docs, but /preflight explicitly notes that /fly subagents do not inherit CLAUDE.md/SessionStart context and only receive docs/ai pre-reading from its own topic map in .agents/skills/preflight/SKILL.md. That map has no long-running/workstream/scratch topic, so multi-session subagents—the scenario this doc targets—will still start without these scratch-control rules. Please add a matching skill/router entry or other activation pointer when registering the doc.
Useful? React with 👍 / 👎.
1 participant
Target\nnew-global-doc: docs/ai/long-running-workstream.md\n\n## Observations addressed\n- Observation 9 (journology/log.md): Long-running self-guided workstreams need a scratch control surface → changed: new doc establishes the scratch.md pattern with Current Truth / Next Steps / Archaeology sections; defines the control-surface vs. evidence-artifact distinction; requires re-read at every compaction/session resume\n- Observation 10 (journology/log.md): Scratch files need compaction-time maintenance rules → changed: new doc adds explicit cleanup pass contract — update Current Truth, move completed items to Archaeology, compact before continuing; adds verification heuristic (if re-reading takes >30 seconds, do a cleanup pass first)\n\nAlso adds the routing table row to
.claude/AGENTS.mdper the new-doc exception:| Long-running multi-session or multi-compaction workstreams | long-running-workstream.md |Generated by Claude Code