菜单
Keeps SESSION-TRAJECTORY.md warm as work progresses and offers checkpoints at natural boundaries. Saving is not stopping.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Review the conversation for threads that were started but not followed — ideas mentioned in passing, decisions deferred implicitly, tensions named but not resolved. Use when taking stock, before saving, or whenever the user wants to know what's been left on the table.
Use when drafting, updating, or reviewing a GitHub pull request title or body, including PR creation flows that need reviewer-facing prose. Follow the repo PR template when present, preserve the user's or project's voice, and route validation/session details to the right surface.
End-of-day close: the user is stepping away and won't serve as a bridge. Full collaborative triage, working-style reflection, and anticipatory framing for the next day. Invoke with /session-close.
Start a new session cycle by reconciling Project Orientation, restoring repo context, and resuming the active bet or opening vision steering when selection genuinely changed.
End-of-cycle transition: interpret the session, triage dangling threads with the user, and prepare SESSION-TRAJECTORY.md for the next cycle. The user is still around and serves as a light bridge. Invoke with /session-rest.
Use when coordinating multiple live Codex threads, maintaining a coordination heartbeat, routing work across implementation threads, inspecting cross-thread ownership or gates, or keeping a coordinated effort moving through PR, review, evidence, and human-action boundaries.
| name | session-save |
| description | Keeps SESSION-TRAJECTORY.md warm as work progresses and offers checkpoints at natural boundaries. Saving is not stopping. |
Saving is not stopping. This skill keeps the trajectory warm so that if the session ends — planned or not — the next session can start with the right direction and momentum. Think of it as version control for context: commit early, commit often, keep working.
Stances used: Load the gap-reading stance for thorough checkpoints, the collaborative-grounding stance for timing decisions, and the relational-continuity stance when writing trajectory updates.
/memories/repo/SESSION-TRAJECTORY.md (accessed via the memory tool) carries two related layers:
Project Orientation is a projection, not a new authority. Reconcile its fields in this order: explicit current project-level user steering; a repo-designated canonical project-state or vision surface; COORDINATION_BRIEF.md together with ACTIVE_PLAN.md; then the latest user-confirmed trajectory as fallback. Apply precedence field by field, let compatible lower sources fill fields a higher source does not address, and record all contributing sources. Within the coordination tier, the brief supplies the slow target and the active plan supplies a compatible Current bet and operational gate; surface a disagreement rather than resolving it by recency. Live workspace evidence can revise proof, route, or gate; it cannot silently replace Vision or Experienced outcome.
As significant work happens during a session, keep SESSION-TRAJECTORY.md loosely updated. This doesn't mean rewriting it after every edit — it means noticing when something meaningful has changed:
When these happen, reconcile Project Orientation first, then update the relevant session-evidence section using the memory tool. This is lightweight — a line or two, not a full transition protocol. Then keep working. Checks and tests are implementation validation; they become Organic proof only when that maintenance or release experience is itself the Experienced outcome.
When the named steering signal arrives, no bet remains, evidence breaks the bet's relationship to the Experienced outcome, or the user explicitly revises project purpose or outcome, the first lifecycle or coordination surface to observe the boundary invokes /steer-by-vision once. Record the transition as steering-in-progress and, when a named signal triggered it, mark that signal reached before handing it off. If the projection already says steering-in-progress, preserve and resume that boundary without another invocation. A completed milestone alone does not reopen selection unless it is the named signal or leaves no active bet.
A checkpoint is a more thorough trajectory capture at a natural boundary. After completing a feature, before starting a risky change, or when enough has accumulated that you'd hate to lose it.
A checkpoint has two parts:
Reconcile and update. Refresh Project Orientation from its named sources, then capture evidence, route changes, decisions, and gotchas around the recorded Immediate gate.
Gather dangling threads. Use the gap-reading stance: read back through the conversation since the last checkpoint and look for what the trajectory update didn't capture. Keep route tensions with the Current bet, preserve already-selected alternatives as Kept-warm arcs with return signals, and reserve project-sized alternatives for the next genuine vision-steering boundary.
The second part is what distinguishes a checkpoint from a quick trajectory update. The trajectory captures the main narrative. Dangling thread review catches what the narrative left out.
After the checkpoint, keep working. A checkpoint preserves progress without costing momentum.
Offer a checkpoint when:
This skill does not run the transition protocol and does not decide whether to stop. When the user decides it's time to transition, they invoke /session-rest (between sessions, user is still around) or /session-close (end of day, user is leaving). This skill is the ongoing maintenance layer; those skills handle transition awareness and the transition protocols.
