registry-sync
Skill0 starsChanged 4 months agoMITView on GitHub
Reconcile OWNERSHIP.md's live-session table from session LOGS, not self-report. Run at the START and END of a /research loop (and any time ownership looks stale) to keep who-owns-what current and avoid two sessions working the same area. Derives each live session's area from its transcript; refreshes timestamps; retires stale (>24h) claims. The automated complement to /mediate.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- gerivdb/SKILLStheir own version, first seen Nov 8, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- gerivdb/SKILLS★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- fleet/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .cursor/rules/atlas.mdcCursor rule
- add-repoSkill
- atlas-setupSkill
- councilSkill
- mediateSkill
- onboardSkill
- optimizeSkill
- researchSkill
- snapshotSkill
- sync-repo-knowledgeSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
START of a /research loop (before picking a task): survey what other live sessions own so you pick an IN-BOUNDS area and don't collide.
- Very specific: line 62
If your intended task is in the OUT-OF-BOUNDS list, it is owned by another live session or a manual claim - pick a different area or …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/registry-sync - keep session ownership current from the logs
The ownership registry goes stale because sessions claim manually (via /mediate) and forget to
update when they finish or switch focus. This skill fixes that by deriving ownership from the ground
truth - the live claude processes and their transcripts - instead of trusting self-report. It is
the automated complement to /mediate (which is for explicit, human-meaningful claims and routing
decisions).
It maintains ONLY the machine-owned "Live sessions (auto-synced)" section of OWNERSHIP.md
(creating it on first run). The hand-curated Areas table above it is read for OUT-OF-BOUNDS context
but never modified.
When to run
- START of a
/researchloop (before picking a task): survey what other live sessions own so you pick an IN-BOUNDS area and don't collide. - END of a
/researchloop (after the closing summary): mark this session's areascomplete. - Any time the registry looks stale or you're about to claim something (
--mode sync).
How - single host (the default for a one-machine setup)
python3 .claude/skills/registry-sync/sync.py --mode start --session "$CLAUDE_CODE_SESSION_ID"
python3 .claude/skills/registry-sync/sync.py --mode end --session "$CLAUDE_CODE_SESSION_ID"
python3 .claude/skills/registry-sync/sync.py --dry-run # preview only, write nothingWhat it does each run (idempotent, safe under concurrent sessions via flock + atomic rename):
- Enumerates live
claudesessions on this host from~/.claude/sessions/*.json(pid alive + procStart matches, to reject pid reuse) -> sessionId + handle (the/renamename). Sessions whose transcript is not under this repo's project dir are ignored. - Derives each session's current area from the tail of its transcript
(
~/.claude/projects/<encoded repo path>/<sid>.jsonl): the most-touchedexperiments/<name>dir. - Rewrites the auto-synced table:
- live session with a row -> refresh
last_active(= transcript mtime); preserve its area/note/claim; un-stale it if it was stale. - live session with no row -> add one, area = its top area dir, note "auto-derived from session log".
- owner not live and
last_active> 24h -> markstatus"stale (session ended)" (reclaimable). --mode end --session <SID>-> mark that session's rowscomplete.
- live session with a row -> refresh
- Prints the live sessions and an OUT-OF-BOUNDS list: areas owned by other live sessions or by
manual Areas-table claims (route to the owner via
/mediate, don't work) vs the rest (IN-BOUNDS).
How - fleet-wide (optional; needs the fleet/ layer)
Home dirs + transcripts are host-local, so one host only sees its own sessions. fleet-sync.sh
pokes every fleet host to emit its snapshot (in parallel), then merges all fresh snapshots into one
registry - a true fleet-wide view:
bash .claude/skills/registry-sync/fleet-sync.sh --mode start --session "$CLAUDE_CODE_SESSION_ID"
bash .claude/skills/registry-sync/fleet-sync.sh --mode end --session "$CLAUDE_CODE_SESSION_ID"Hosts come from $ATLAS_FLEET_HOSTS (space-separated) or the node names in fleet/fleet.yaml.
The repo must be checked out (or mounted) at the same path on every host. Down/unreachable hosts
are skipped; their last snapshot ages out via the 2h snapshot-freshness rule, then the 24h
last_active rule. Snapshots land in .atlas/ownership/hosts/ (gitignored).
Acting on the output
- If your intended task is in the OUT-OF-BOUNDS list, it is owned by another live session or a
manual claim - pick a different area or coordinate (per
/mediate: route, don't halt). - For explicit, non-experiment-dir claims (a cross-cutting refactor, the manuscript), use the
hand-curated Areas table in
OWNERSHIP.md(via/mediate); this skill leaves that table alone and folds its claims into the OUT-OF-BOUNDS report.
Ownership invariant (keep this true)
A sync run may only mark ITS OWN CALLER's rows terminal (--mode end -> complete for
owner_session == --session). It must never set another session's areas to a terminal/released
state, never delete another session's row, and never edit the manual Areas table. It may only, for
other sessions: refresh last_active, un-stale a live row, add an auto-derived row, or mark a row
stale once last_active > 24h. This invariant is what keeps fleet-sync.sh in-bounds to run
against the whole fleet without per-owner mediation. If you ever change reconcile() so a run can
terminate another session's work, that call becomes OUT-OF-BOUNDS and must route to that owner
first (/mediate).
Configuration (env overrides)
ATLAS_REPO- the governed repo (default: git toplevel of the cwd, else the skill's own repo).ATLAS_OWNERSHIP- registry file (default<repo>/OWNERSHIP.md).ATLAS_PROJECTS_DIR- transcripts dir (default~/.claude/projects/<encoded repo path>).ATLAS_HOSTS_DIR- fleet snapshot dir (default<repo>/.atlas/ownership/hosts).ATLAS_AREA_RE- regex for area detection (defaultexperiments/([A-Za-z0-9][A-Za-z0-9_-]*)).ATLAS_FLEET_HOSTS- space-separated host list forfleet-sync.sh(elsefleet/fleet.yaml).
Per-host note
Reads this host's ~/.claude; the registry in the repo is shared (same checkout/mount, or synced
via git). Each host's runs reconcile their own sessions; the 24h-stale rule retires sessions from
hosts that have gone quiet.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

