القائمة
End-of-session procedure that captures learnings into memory topic files, verifies index consistency, updates current-state.md, and syncs changes to team-lib.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Send an on-demand manual pulse to the ClickUp Pulse channel and optionally register thread-to-task mappings.
Graduated to team-lib — see team-lib/skills/session-debrief/. This stub demonstrates the my-lib → team-lib graduation pattern.
Add aliases, functions, or env vars to ~/.bashrc with section-aware placement
Check and upgrade local workspace files against latest team-lib templates
Compose base + type-override templates with brand tokens to produce fully resolved branded templates. Deep merges base+override, resolves tokenRef dot-paths against brand-tokens.json, and applies brand preference fan-out.
Create or update a brand-guidelines.md file for any brand — the single source of truth for all branded content generation.
| name | session-debrief |
| description | End-of-session procedure that captures learnings into memory topic files, verifies index consistency, updates current-state.md, and syncs changes to team-lib. |
| summary | Run at end of a work session to persist learnings, check index health, update workspace current state, and propagate improvements to team-lib. Uses preflight/postflight scripts for speed and reliability. |
| version | 2.4.0 |
| created | "2026-02-20T00:00:00.000Z" |
| last_updated | "2026-05-04T00:00:00.000Z" |
| maintainer | pvragon |
Run this skill at the end of a work session to capture learnings and maintain workspace hygiene.
Architecture: Deterministic work is pushed to shell scripts (preflight.sh, postflight.sh). The LLM focuses only on judgment-requiring work (memory, state updates, summaries). This makes the debrief faster and more reliable.
Run the preflight script to collect all session data in one pass:
bash ~/ai-workspace/team-lib/skills/session-debrief/preflight.sh
This outputs a JSON report with:
git_changes — files changed, recent commits, which registered dirs were touchedregistry_issues — files missing from registries or registry entries pointing to missing filessync_needed — AGENTS.md or skills that changed and also exist in team-libstale_flags — current-state.md health (stale entries, pending notes, old decisions)memory_issues — MEMORY.md index inconsistencies (unindexed files, broken links)session_info — date, approximate timestamps, memory index line count, and a session_marker token you must pass to postflightRead the report carefully before proceeding. It tells you exactly what needs attention.
Grab session_info.session_marker — you'll pass it to postflight so it can deterministically identify the current session's JSONL (needed for setting the /resume title correctly when multiple concurrent sessions exist).
With the preflight report in hand, work through these items. Do them all before moving to Phase 3.
If registry_issues is non-empty, fix each one:
unregistered — Add the file to the appropriate registry/*.yamlmissing_file — Remove the stale entry from the registryIf memory_issues is non-empty:
unindexed — Add the file to MEMORY.md with a one-line summarymissing_file — Remove the broken link from MEMORY.mdReflect on the session and answer these questions. If the answer is yes, act immediately:
Learned something future agents should know?
→ Write to the appropriate topic file in memory. If no existing topic fits, create a new one and add a row to MEMORY.md.
Discovered a convention or gotcha?
→ Append to lessons-learned.md in the memory directory.
Learned about a person, process, or entity? → Find or create the relevant topic file.
memory/current-state.md is a structured snapshot, not an append log. It holds only the four sections below. Verbose session detail belongs in memory/short-term/YYMMDD-facts.md (Phase 2c — and Move 1 of the 4-tier memory architecture).
🚫 DO NOT append "## Session Add" blocks to current-state.md. This was the historical bloat pattern that produced a 228KB file. Update sections in place — replace stale content, don't accrete it. (Reset 2026-05-04: prior session-add history → current-state-archive.md.)
Open memory/current-state.md and update:
Last updated (or Last reset) header comment.Size discipline: Target ≤5KB total. If you find yourself wanting to add a long session narrative, that's a signal it belongs in short-term/YYMMDD-facts.md instead. The structured sections here distill current state, not session history.
Cleanup (informed by preflight stale_flags):
pending_notes → Previous notes should have been consumed. Clear stale ones.old_decisions → Prune entries older than 14 days. Move long-term-worthy decisions to memory topic files first.If sync_needed is non-empty:
agents_md → Run the push-agents-to-template skill.skill:<name> → Review whether the skill changes should propagate to team-lib. Only promote stable, generally-useful changes (not workspace-specific tweaks).If nothing in sync_needed, skip this.
Append a one-line entry to memory/session-log.md:
Format: YYYY-MM-DD | session-name | key topics/outcomes
/rename name if set, otherwise compose a 2-4 word descriptive name<!-- entries below -->)/renameWrite a 1-3 sentence debrief message. Do not post it yet — it will be passed to postflight.
Format:
JH Claude Debrief [<start time> - <end time>]: <summary of outcomes, decisions, and next steps>
Use session_info.approx_start and session_info.end from the preflight report for timestamps.
Before invoking postflight, build two explicit space-separated lists of repo-relative paths the debrief modified — one for the my-lib repo, one for the agents identity repo. Postflight will refuse to run without these. This replaces the old git add -A catch-all behavior (which previously swept up unrelated work from concurrent sessions — see commit 836c2c6 and the 2026-04-20 21:11 incident in current-state.md).
What counts as "touched":
current-state.md, session-log.md, MEMORY.md, etc.)cp, mv, hook scripts, etc.) — these don't show in tool-call history; track them deliberately as you gogit commit on a file, don't pass it again; postflight won't have anything to add.What does NOT count:
git_changes.changed_files baseline). Those belong to whatever produced them (likely a concurrent session) and must be preserved as-is.Path format: repo-relative (e.g., backlog/foo.md, memory/MEMORY.md), space-separated.
If nothing was touched in a given repo, pass an empty string (--mylib-files "") to skip that repo's commit cleanly.
Run the postflight script with the composed pulse message and the file lists from Phase 2h:
bash ~/ai-workspace/team-lib/skills/session-debrief/postflight.sh \
--session-name "<session-name>" \
--session-marker "<value from preflight's session_info.session_marker>" \
--mylib-files "backlog/foo.md skills/bar/SKILL.md" \
--agents-files "memory/MEMORY.md memory/topic_xyz.md" \
--pulse-message "<debrief message>"
IMPORTANT — required args (will error if missing):
--mylib-files "..." — explicit space-separated paths the debrief modified in my-lib (or "" to skip)--agents-files "..." — explicit space-separated paths the debrief modified in agents identity repo (or "" to skip)IMPORTANT: Pass --session-marker from the preflight report's session_info.session_marker. Postflight will grep JSONLs for that marker to identify the current session deterministically — no guessing, safe with any number of concurrent sessions. (--session-id <uuid> is still accepted as an explicit override if you already know the UUID.)
This handles in one pass:
*:Zone.Identifier junk files across ~/ai-workspace (via executions/clean_zone_identifiers.sh)--agents-files)--mylib-files)Title logic: If the user did /rename, uses that name verbatim. Otherwise, prepends YYMMDD-HH:MM <session-name> to the JSONL head so /resume shows a meaningful name.
Options:
--mylib-files "<paths>" — REQUIRED unless --skip-commit or --legacy-add-all. Space-separated repo-relative paths to stage for the my-lib commit. Pass "" to skip that repo cleanly.--agents-files "<paths>" — REQUIRED unless --skip-commit or --legacy-add-all. Same shape as --mylib-files, for the agents identity repo.--legacy-add-all — Escape hatch: revert to the old git add -A catch-all behavior. Use only when explicitly justified (e.g., recovering after extensive uncommitted work that genuinely all belongs to this debrief). Logs a prominent warning. Prefer explicit file lists.--session-marker "<token>" — Marker from preflight output; resolves to the current session's UUID via JSONL grep (preferred)--session-id "<uuid>" — Explicit session UUID (skip marker lookup; use only if you already know it)--skip-pulse — Skip posting (e.g., if the user declines)--skip-commit — Dry run (adapters only, no git commits — also bypasses the file-list requirement)Summarize to the user what was updated: