sibling-sync

name	sibling-sync
description	Bilateral SYNERGY/UPSTREAM reconciliation across sibling projects. Use when the user wants to sync sibling SYNERGY/UPSTREAM files, compare both sides to surface drift, find reciprocation gaps (entries here but not there, or vice versa), flag stale-aligned rows, detect status drift across sides, surface friction the sibling tracks ABOUT this project (their UPSTREAM-<this-project>.md), or apply a reciprocation batch with --auto-reciprocate. Workflow 3 covers two UPSTREAM pairing modes: shared third-party dependencies AND reciprocal sibling-friction pairs (UPSTREAM-<sibling>.md here ↔ UPSTREAM-<this-project>.md there). NOT for logging entries on this side (use /synergy-tracker workflow 1 (Log a synergy entry)) — sibling-sync compares both sides without writing by default. NOT for upstream → project drift (use /vendor-sync); sibling-sync handles peer-to-peer drift between sibling vp-* projects. Trigger phrases: 'sibling sync', 'compare siblings', 'sync sibling', 'reconcile siblings', 'reciprocation gap', 'sync drift', 'bilateral sync', 'sync SYNERGY', 'sync UPSTREAM both ways', 'auto-reciprocate', 'check sibling drift', 'peer-to-peer drift', 'cross-project drift', 'sibling reconciliation', 'sibling has friction about us', 'what does the sibling say about us', 'reconcile sibling-tracked friction', 'reciprocal upstream friction', 'friction filed against this project', 'sibling follow-up', 'act on sibling drift', 'after sibling-sync', 'follow-up actions', 'what to do about sibling findings'.
user-invocable	true
argument-hint	[--auto-reciprocate] [sibling-name]
paths	["SYNERGY-*.md","UPSTREAM-*.md",".claude/synergy-registry.json",".claude/vendor-registry.json"]
allowed-tools	["Bash","Read","Glob","Grep","Edit","Write","Skill","AskUserQuestion","mcp__basic-memory__search_notes","mcp__basic-memory__read_note"]

Sibling Sync

Bilateral reconciliation of SYNERGY-*.md and UPSTREAM-*.md files between this project and its sibling vp-* projects. Read-only by default — surfaces drift, reciprocal gaps, stale-aligned rows, and status drift across sides without mutating anything. The opt-in --auto-reciprocate flag writes reciprocal entries to the sibling's SYNERGY file via per-entry confirmation.

Companion to /vendor-sync (which handles upstream → project drift); this skill handles peer-to-peer drift between siblings registered in .claude/synergy-registry.json.

Design Rationale

The bd v1.0.0 Integration Charter (gastownhall/beads@5d524cf7:docs/INTEGRATION_CHARTER.md) explicitly punts cross-tracker orchestration out of bd's scope: bd will never grow a feature that routes a cross-project item from project A's tracker to project B's tracker. /sibling-sync is exactly the workflow-automation layer the Charter defers to external tools — file-based reconciliation between sibling vp-* projects, mediated by registries and confirmation prompts rather than synchronous tracker calls.

This mirrors the rationale already cited by /synergy-tracker for keeping cross-project state in SYNERGY-*.md plus Basic Memory rather than in bd.

Cross-skill boundaries

/sibling-sync is a comparison and reconciliation layer that sits alongside the per-side logging skills. It owns nothing in Basic Memory and nothing on this project's side of the SYNERGY/UPSTREAM files.

• Does NOT write SYNERGY entries on this project's side. /synergy-tracker workflow 1 (Log a synergy entry) owns logging on this side.
• Does NOT pull upstream subtrees. /vendor-sync owns subtree pulls and the upstream → project drift workflow.
• Does NOT write Basic Memory notes. /synergy-tracker workflow 5 (Promote to Basic Memory) owns ## Cross-Project Synergy writes to sibling entity notes; /upstream-tracker workflow 6 (Promote to Basic Memory) owns ## Upstream Friction writes. Basic Memory write tools are intentionally absent from this skill's allowed-tools.
• Does NOT write ## Trend Reviews entries to SYNERGY files. Those belong to /synergy-tracker workflow 4 (Trend review (quarterly)). Even under --auto-reciprocate, /sibling-sync only mirrors content entries into reciprocal sections — never trend-review summaries.
• Stale-row detection is INLINE here for the threshold values used during comparison runs. The canonical staleness-threshold definition lives in /synergy-tracker workflow 4 (Trend review (quarterly)) — workflow 2 (Sync sibling SYNERGY) below cites it. Per RETRO-10 YAGNI guard: extract this to a shared helper only when a third skill needs the same logic.
• Surfacing reciprocal-friction findings is in scope; acting on them is not. Workflow 3 (Sync sibling UPSTREAM) Mode B (see below) reads the sibling's UPSTREAM-<this-project>.md to surface friction the sibling tracks about this project. Filing the resulting work as bugs/features/opportunities on this side is /upstream-tracker workflow 1 (Log a new entry)'s job. Annotating the sibling's entry as resolved is /upstream-tracker workflow 3 (Resolve an entry)'s job, performed on the sibling's side. /sibling-sync reports only.
• Orchestrator role for follow-up actions (v0.14.0). Workflows 2 (Sync sibling SYNERGY) and 3 (Sync sibling UPSTREAM) end with a per-sibling action menu (see "Action-menu protocol" below) that delegates writes to the owning skill (/vp-beads:synergy-tracker, /vp-beads:upstream-tracker) via the Skill tool, or runs bd create directly for beads issues. /sibling-sync still owns nothing in Basic Memory and nothing in this project's SYNERGY/UPSTREAM files — ownership boundaries are unchanged.

Registry and path resolution

Sibling projects are declared in .claude/synergy-registry.json (array of {name, file, remote, bm-entity, relationship, local-path?} entries). The optional local-path field gives the on-disk path to the sibling checkout (relative paths resolve from this project root). When absent, fall back to ../<name>/.

.claude/synergy-registry.local.json is a gitignored companion that overrides fields in the committed registry — same per-entry merge by name pattern as .claude/vendor-registry.local.json. Resolution order:

1. Read .claude/synergy-registry.json.
2. If .claude/synergy-registry.local.json exists, merge it on top by name key. Fields in .local.json win; absent fields keep the base value. Entries in .local.json whose name is not in the base registry are ignored.
3. For each merged entry, resolve local-path (registry value or ../<name>/).
4. If the resolved path does not exist on disk, report informatively and SKIP that sibling. Do not error out — continue with siblings that are accessible.

Workflow 3 (Sync sibling UPSTREAM) additionally consumes the merged .claude/vendor-registry.json (+ .local.json) to identify shared vendor dependencies across siblings.

Workflows

Determine which workflow the user needs based on their request. If ambiguous, default to running workflow 1 (Discover sibling(s)) followed by workflow 2 (Sync sibling SYNERGY) and workflow 3 (Sync sibling UPSTREAM) as a single report. Workflow 4 (Apply reciprocation batch) only fires under explicit --auto-reciprocate.

1. Discover sibling(s)

Resolve which siblings will participate in this run.

Steps:

1. Read .claude/synergy-registry.json. If the file does not exist, redirect: tell the user no sibling registry is configured and offer to invoke /synergy-tracker workflow 1 (Log a synergy entry), which will run the guided registry creation flow at step 1b for the first sibling. If the user names a sibling now, follow the synergy-tracker step 1b prose from this conversation (re-reading skills/synergy-tracker/SKILL.md workflow 1 (Log a synergy entry) step 1b and applying its logic in-session — Claude Code has no actual inline-skill-invocation mechanism, so this means executing step 1b's instructions verbatim from sibling-sync's context). After the registry is created, resume from step 2 below. Otherwise stop, and instruct the user to invoke /synergy-tracker directly with the sibling name and then re-run /sibling-sync.
2. If .claude/synergy-registry.local.json exists, merge it on top per the per-entry merge rules in the Registry section above.
3. If the user named a specific sibling in their request or argument, filter the merged list to that entry. Otherwise, all merged entries participate.
4. For each entry, resolve local-path → ../<name>/ fallback. Probe each resolved path with a directory existence check.
5. Build two lists:
   • Accessible siblings (path exists) — proceed to workflow 2 (Sync sibling SYNERGY) and workflow 3 (Sync sibling UPSTREAM) for each
   • Inaccessible siblings (path missing) — report them so the user knows what was skipped, with the resolved path and a hint that .claude/synergy-registry.local.json can override the path
6. Report the participation list before continuing.

Output:

Siblings participating:
- vp-knowledge → /Users/.../vp-claude  (registry local-path)

Siblings skipped (path not accessible):
- vp-other → ../vp-other  (set local-path in synergy-registry.local.json)

If no siblings are accessible, stop and report. The user can either correct the paths via .claude/synergy-registry.local.json or accept that this run has no work to do.

2. Sync sibling SYNERGY

For each accessible sibling from workflow 1 (Discover sibling(s)), compare the bidirectional SYNERGY files and surface drift findings. Report only — no writes.

Steps:

1. Read this project's SYNERGY-<sibling>.md. If absent, treat as zero entries and proceed (the gap will surface as "Unreciprocated entries on sibling" if the sibling has any entries).

2. Read the sibling's SYNERGY-<this-project>.md at <resolved-local-path>/SYNERGY-<this-project>.md. If absent, treat as zero entries.

3. Parse each side's entries section-by-section (Shared Patterns, Divergences, Extraction Candidates, They Have / We Don't). Build a bidirectional entry map keyed by title, normalized per the rule in Guidelines below (2-pass matching: deterministic lead-clause pass, then judgment pass on residuals).

   Section migration is its own signal — not silently merged. Matching is within-section: an entry's title is paired only with same-section titles on the sibling. If an entry has migrated sections on one side (e.g., Shared Pattern here, Divergence on sibling — typical when one side promoted after noticing drift), it surfaces as (a) on the originating section and (b) on the destination section, NOT as a finding (d) Status drift. The section migration is itself the drift signal worth surfacing, and the (a)/(b) framing tells the user precisely which side moved. Finding (d) applies within-section only.

   Section asymmetry — excluded from findings (a)/(b): the They Have / We Don't section is intrinsically asymmetric. Entries here describe what the sibling has that we don't; reciprocally on the sibling's side, the same-named section describes what we have that they don't — a different semantic set. Bilateral title comparison is meaningless for this section. Skip its entries when computing findings (a) and (b). The user can read the section directly to act on adoption candidates; logging an adoption decision is /synergy-tracker's job, not /sibling-sync's.

4. Walk the merged map and classify each entry into one of four findings:

   • (a) Reciprocal gaps — entries on this side with no matching title on the sibling. The sibling lacks the reciprocal entry. Candidates for workflow 4 (Apply reciprocation batch) under --auto-reciprocate. Excludes entries from They Have / We Don't (asymmetric — see step 3).

   • (b) Unreciprocated entries on sibling — entries on the sibling with no matching title here. The user may want to invoke /synergy-tracker workflow 1 (Log a synergy entry) to log these on this side. /sibling-sync does NOT write to this side automatically. Excludes entries from They Have / We Don't (asymmetric — see step 3).

   • (c) Stale alignment claims — entries with Status: aligned and Last verified: more than 8 sprints old (≈ two trend-review cycles). Inline threshold; canonical definition is /synergy-tracker workflow 4 (Trend review (quarterly)). Treat 1 sprint ≈ 2 weeks if no other calibration is available; if the entry has no Last verified: field, fall back to the entry's date stamp.

   • (d) Status drift — matched entries whose Status: field differs across sides. Applies to two cases:

     1. Shared Patterns where one side records aligned and the other records drifting or diverging (or any disagreement on the Status value). Often signals that one side has converged or re-diverged without the reciprocal note being refreshed.
     2. Divergences with Convergence path: adopt-theirs or Convergence path: propose-shared where one side has moved to adopted/converged while the other still says drifting or similar.

     Excludes Divergences with Convergence path: accept-difference — those are intended-asymmetric and have no drift signal to flag.

5. Output a structured report grouped first by sibling, then by finding category. Include each entry's title, both sides' values where they differ, and a one-line action hint per category.

Output format (per sibling):

## SYNERGY drift — vp-knowledge

### (a) Reciprocal gaps (here, missing on sibling)
- "Hook event coverage" (Shared Patterns) — sibling has no matching entry
  → /sibling-sync --auto-reciprocate to file the reciprocal

### (b) Unreciprocated entries on sibling
- "validate-plugin tool-reference audit" (Shared Patterns) — we don't track this
  → /synergy-tracker to log on this side

### (c) Stale alignment claims (>8 sprints since Last verified)
- "PreCompact prompt command hook" — Last verified: 2026-01-15 (here),
  2026-01-15 (sibling). Re-verify now.

### (d) Status drift
- "npm-run-all2 parallel check stages" (Shared Patterns) — Status here:
  diverging. Status sibling: aligned. Sibling converged; refresh this row.
- "BM section ownership scheme" (Divergences, propose-shared) — Status
  here: drifting. Status sibling: aligned.

6. Offer follow-up actions. After workflow 3 (Sync sibling UPSTREAM) has also finished printing for this sibling, prepare the SYNERGY tier of the action menu (built from this workflow's findings) and let workflow 3 (Sync sibling UPSTREAM) step 6 pair it with the UPSTREAM tier into a single AskUserQuestion call. Do NOT issue the prompt from this step — the prompt is dispatched from workflow 3 (Sync sibling UPSTREAM) step 6 once both reports are on screen. Full protocol — which findings map to which menu options, the header values, the --auto-reciprocate precedence rule, and the plugin-namespaced Skill invocations — lives in the "Action-menu protocol" section below.

3. Sync sibling UPSTREAM

For each accessible sibling, build two kinds of UPSTREAM file pairs and compare friction tracking on each. Same report-only contract as workflow 2 (Sync sibling SYNERGY).

Two pairing modes coexist; both can fire on a single sibling:

• Mode A — shared-dependency pairing. Both sides have UPSTREAM-<dep>.md with the same basename (e.g., both have UPSTREAM-basic-memory.md). The files describe the same third-party dependency <dep>. Findings (a)–(d).
• Mode B — reciprocal sibling-friction pairing. This project has UPSTREAM-<sibling-name>.md (friction we log about the sibling) AND/OR the sibling has UPSTREAM-<this-name>.md (friction the sibling logs about us). Different basenames; same bilateral relationship. Owner-side semantics invert relative to Mode A: an entry in the sibling's UPSTREAM-<this-name>.md with Ownership: upstream means THIS project is the upstream that must act. Findings (e)–(h).

Steps:

1. Build Mode A pairs. Glob this project for UPSTREAM-*.md. Glob the sibling's resolved local-path for UPSTREAM-*.md. Compute the intersection by basename — each match is one Mode A pair. Record both sides' full UPSTREAM basename lists for use in step 2.

2. Detect Mode B pair. Derive this project's canonical name per the four-tier algorithm in skills/synergy-tracker/references/project-name-derivation.md to compute <this-name>. Apply the same algorithm (tier 3 for the sibling subject) to the registry name field for <sibling-name>. Stale local-path guard: if the registry entry specifies a local-path that does not resolve to an accessible directory, tier 1 (sibling-registry back-pointer) silently falls through to tier 2 (this project's plugin.json). Warn the user before falling through: "Sibling local-path is not accessible — tier 1 derivation skipped; <this-name> may diverge from how the sibling registered this project. Update .claude/synergy-registry.local.json if the sibling moved." Then check:

   • Does the sibling have <resolved-local-path>/UPSTREAM-<this-name>.md?
   • Does this project have UPSTREAM-<sibling-name>.md?

   If either file exists, this sibling has a Mode B pair (one-sided or two-sided). Both files absent is normal — no reciprocal friction tracked on either side. Skip Mode B for this sibling and continue.

   The Mode B file pair is <this-root>/UPSTREAM-<sibling-name>.md ↔ <sibling-root>/UPSTREAM-<this-name>.md. By construction these basenames differ from any Mode A pair's basename (Mode A keys on shared third-party dep names; Mode B keys on sibling project names that appear in the synergy registry). No deduplication guard needed.

3. Process Mode A pairs. For each Mode A pair, read both copies and parse entries (Bugs, Feature Requests, Upstream Opportunities, Resolved). Build a bidirectional entry map by title using the same 2-pass matching rule as workflow 2 (deterministic lead-clause Pass 1 + judgment Pass 2 on residuals — see Guidelines). UPSTREAM titles are typically more structured than SYNERGY titles, so Pass 2 fires less often, but the rule is identical for consistency. Classify each entry:

   • (a) Duplicate friction — same title on both sides. Sanity check: are the workarounds, dates, and status fields aligned? If not, it's a candidate for category (b).
   • (b) Complementary workarounds — same title both sides but the Workaround: field (or equivalent) differs. The sibling may have found a better mitigation. Flag for cross-pollination.
   • (c) Stale entries — entries dated more than 3 months ago without a Trend Review annotation since. Either side. Stale ≠ wrong, but worth re-verifying.
   • (d) Sibling-only entries — friction the sibling tracks for a shared dependency that we don't. Potential adoption: invoke /upstream-tracker workflow 7 (Sync from Basic Memory) or workflow 1 (Log a new entry) to bring matching entries over here. /sibling-sync does NOT write here automatically.

4. Process Mode B pair. Read whichever side(s) of the pair exist. Parse entries the same way as step 3. Match titles bidirectionally with the same 2-pass rule. Classify each entry:

   • (e) Sibling's unresolved friction against this project — entries in <sibling-root>/UPSTREAM-<this-name>.md that are NOT prefixed with _(Resolved ...)_ and NOT in a ## Resolved section if one exists. Ownership: upstream on these entries means THIS project owns the fix (we are upstream from the sibling's perspective). Surface ALL unresolved entries — every one is a request directed at us. Action hint: file beads issues here or address inline; consider logging a cross-reference in our UPSTREAM-<sibling-name>.md if a workaround is built.
   • (f) Our unresolved friction against the sibling — entries in <this-root>/UPSTREAM-<sibling-name>.md that are unresolved on our side and have no corresponding _(Resolved ...)_ annotation on either side. Informational: documents work blocked on the sibling. Action hint: check sibling release notes or changelog for shipped fixes the sibling forgot to annotate.
   • (g) Cross-side staleness — our entry, sibling may have shipped. Entry in <this-root>/UPSTREAM-<sibling-name>.md unresolved on our side, but the sibling shows a "shipped" signal (see "What 'shipped' means" below). Use 6 months as the look-back horizon for git-log scanning. Action hint: re-verify against the sibling's current release; annotate with _(Resolved ...)_ via /upstream-tracker workflow 3 (Resolve an entry) if confirmed shipped.
   • (h) Reverse cross-side staleness — sibling tracks us, we may have shipped. Entry in <sibling-root>/UPSTREAM-<this-name>.md unresolved on the sibling's side, but this project shows a "shipped" signal (recent CHANGELOG entry, _(Resolved ...)_ in our cross-reference, or git tag/commit subject within 6 months matching the entry title). Read-only finding: /sibling-sync cannot write the sibling's file. Action hint: notify sibling maintainer, or raise on their side via /upstream-tracker workflow 3 (Resolve an entry) so they can annotate.

   What "shipped" means (pinned definition for findings (g) and (h)): a fix is shipped when (1) a CHANGELOG or _(Resolved ...)_ annotation exists on the owner's side, OR (2) the feature/fix is referenced in a git tag message or commit subject within the relevant release window (use git -C <owner-path> log --oneline --since="6 months ago" as a heuristic proxy — string-match the entry title or its lead clause; do not parse). A Workaround: full on the filing side without a corresponding shipped version on the owner's side is NOT sufficient; that is the filing project's mitigation, not upstream resolution.

5. Output. Report Mode A findings first (grouped by sibling, then by shared dependency, then by finding category), then Mode B findings (grouped by sibling) under a separate header. This ordering keeps the existing Mode A output shape intact and adds Mode B as an additive block.

Note on UPSTREAM coverage gaps: /sibling-sync now handles two cases: shared third-party dependencies (Mode A, basename intersection) and reciprocal sibling-friction pairs (Mode B, inverse-name detection). One-sided UPSTREAM files about non-sibling, non-shared dependencies are still out of scope — those are the sibling's responsibility to discover via /upstream-tracker workflow 7 (Sync from Basic Memory) on its own.

Output format additions for Mode B:

## UPSTREAM reciprocal-friction — vp-knowledge

(Mode B: this-side UPSTREAM-vp-knowledge.md ↔ sibling-side UPSTREAM-vp-beads.md)

### (e) Sibling's unresolved friction against this project (we should action)
- "vp-beads: new /sibling-sync skill" (Feature Requests, 2026-05-04) — sibling
  marks Workaround: partial; we shipped in v0.12.0. See finding (h).
- "synergy-tracker: mandate bilateral reciprocation" (Feature Requests, 2026-05-04)
  Ownership on their side: upstream (us) · Workaround on their side: full
  → file beads issue or address inline

### (f) Our unresolved friction against the sibling
- "Agent effort defaults not overridable from parent" (Feature Requests, 2026-04-05)
  Ownership: upstream (them) · Workaround: none

### (g) Cross-side staleness: our entry the sibling may have shipped
- (none this run)

### (h) Reverse staleness: sibling tracks us but we may have shipped
- "vp-beads: new /sibling-sync skill" — v0.12.0 tag (2026-05-05) matches.
  Sibling should annotate _(Resolved 2026-05-05, vp-beads v0.12.0)_.
  → notify sibling maintainer; cannot write their file from here

6. Offer follow-up actions. This is the dispatch point. After this workflow finishes printing (and workflow 2 (Sync sibling SYNERGY) has already finished for the same sibling), build a single AskUserQuestion call combining workflow 2 (Sync sibling SYNERGY)'s SYNERGY tier with this workflow's UPSTREAM tier. The "Action-menu protocol" section below specifies the full UPSTREAM tier: findings (b) and (d) collapse into a single "Update our UPSTREAM" option that delegates to /vp-beads:upstream-tracker; finding (e) routes directly to bd create; finding (g) delegates to /vp-beads:upstream-tracker workflow 3 (Resolve an entry). Findings (a), (c), (f), and (h) are informational and not present in the menu. If no UPSTREAM findings are actionable AND no SYNERGY findings are actionable for this sibling, skip the prompt for this sibling. If only one tier has actionable findings, issue a single-question call.

4. Apply reciprocation batch

Opt-in mutation path. Only runs when the user supplies --auto-reciprocate in their invocation, or explicitly confirms intent like "yes, apply all the reciprocal gaps". Mirrors /upstream-tracker workflow 7 (Sync from Basic Memory)'s per-entry confirmation pattern.

Steps:

1. Re-run workflow 2 (Sync sibling SYNERGY) finding (a) (reciprocal gaps) for each accessible sibling, applying the stricter matching rules from the Hard Limits section below: Pass 1 (deterministic) matches only; any Pass 2 (judgment) matches from workflow 2 (Sync sibling SYNERGY) are added back to the reciprocation queue with an extra disambiguation prompt rather than suppressed silently. These are the entries on this side that the sibling demonstrably lacks.
2. For each reciprocal gap, in order:
   1. Read the source entry from this project's SYNERGY-<sibling>.md (full entry text including title, date, structured fields).
   2. Determine the destination file at the sibling: <resolved-local-path>/SYNERGY-<this-project>.md (derive <this-project> per skills/synergy-tracker/references/project-name-derivation.md). If it does not exist yet, plan to Write a new file using the four-section template from skills/synergy-tracker/references/synergy-entry-format.md.
   3. Determine the destination section from the source entry's section (a Shared Pattern on this side becomes a Shared Pattern on the sibling, etc.).
   4. Show the user: source entry text + destination file path + destination section. Ask: "Write reciprocal entry to <sibling-path>/SYNERGY-<this-project>.md under ### <Section>? [y/n/skip-rest]".
   5. On y: append the entry under the destination section using Edit (or Write if the file is new). Replace any _No entries yet._ placeholder in that section with the entry. Keep the entry text as-is from this side — do not rewrite to the sibling's voice; reciprocation IS the verification step (per /synergy-tracker workflow 1 (Log a synergy entry) bilateral mandate). The sibling will re-verify on their next reciprocation pass.
   6. On n or skip-rest: skip and continue (or stop the batch on skip-rest).
3. After the batch, report:
   • Entries written, with destination file paths
   • Entries skipped, with reason
   • Verification reminder for the user: run git status in the sibling repo, review the appended entries, commit on that side. /sibling-sync does not commit on the sibling's behalf. Also remind the user to file a beads follow-up on the sibling for re-verification next sprint, per /synergy-tracker workflow 1 (Log a synergy entry)'s reciprocation mandate.

Hard limits on workflow 4 (Apply reciprocation batch):

• Only mirrors entries from workflow 2 (Sync sibling SYNERGY) finding (a). Does NOT mirror UPSTREAM entries from workflow 3 (Sync sibling UPSTREAM) — /upstream-tracker workflow 7 (Sync from Basic Memory) is the right channel for cross-project UPSTREAM adoption (BM is the cross-project bridge for friction; SYNERGY is the cross-project bridge for patterns). This applies equally to Mode A findings (a)–(d) AND Mode B findings (e)–(h): finding (e) entries get filed natively on this side via /upstream-tracker workflow 1 (Log a new entry), not mirrored; finding (h) annotations get written by the sibling via their own /upstream-tracker workflow 3 (Resolve an entry), not by us.
• Never mirrors entries from ## They Have / We Don't. The section is intrinsically asymmetric (entries here describe sibling capabilities WE lack; the sibling's same-named section describes the inverse asymmetry). Workflow 2 (Sync sibling SYNERGY) already excludes this section from finding (a), but this is restated here as a mutation-side guard: even if a future edit relaxes the workflow 2 (Sync sibling SYNERGY) exclusion, workflow 4 (Apply reciprocation batch) must never write a They Have / We Don't entry to the sibling.
• The reciprocal-gap list is computed using Pass 1 matches only. Entries that paired via Pass 2 (judgment) in workflow 2 (Sync sibling SYNERGY) are added back to the reciprocation queue and presented to the user with a flag: "This entry may already exist on the sibling as <pass-2-matched-title> — does that match? [y=skip / n=write reciprocal anyway / skip-rest]". Defaulting to caution at the mutation boundary inverts the read-only cost asymmetry: under --auto-reciprocate, suppressing a write that should happen (false-positive Pass 2 match) is more expensive than proposing a duplicate the user can reject (false-negative).
• Never writes to ## Trend Reviews sections on either side.
• Never writes to this project's side. Reciprocal entries go to the sibling only — logging on this side is /synergy-tracker workflow 1 (Log a synergy entry)'s job.
• Never writes to Basic Memory (no BM edit tooling allowed in this skill). BM writes are /synergy-tracker workflow 5 (Promote to Basic Memory) and /upstream-tracker workflow 6 (Promote to Basic Memory)'s territory.

Action-menu protocol

This skill never writes SYNERGY/UPSTREAM/BM directly — even the menu options dispatch to the owning skill via the Skill tool, or run bd create for beads issues. The menu is a navigation aid, not a write path. The default read-only contract from earlier versions still holds: a user who picks "None" for both questions receives the report and exits without any mutation.

After workflows 2 (Sync sibling SYNERGY) and 3 (Sync sibling UPSTREAM) have printed their per-sibling reports, sibling-sync issues a single AskUserQuestion call with up to two single-select questions per sibling. The AskUserQuestion SDK contract caps options at 2-4 per question (plus an auto "Other"); we therefore split SYNERGY and UPSTREAM into separate questions rather than one flat menu. Skipping a question is just selecting its "None" option; both tiers default to read-only on skip.

Two-tier menu shape

Q1 — SYNERGY follow-up (header: "Synergy", 7 chars). Options listed only when their finding count is nonzero:

Option	Trigger	Dispatch
1. Apply reciprocal gaps (N)	finding (a) > 0	re-enter workflow 4 (Apply reciprocation batch) in-skill — no Skill call
2. Log unreciprocated sibling entries (N)	finding (b) > 0	Skill(skill="/vp-beads:synergy-tracker", args=...) → workflow 1 (Log a synergy entry)
0. None — synergy report only	always	exit SYNERGY tier without action

Q2 — UPSTREAM follow-up (header: "Upstream", 8 chars). Options listed only when their finding count is nonzero. Findings (b) and (d) collapse into a single option to keep the question within the 4-option SDK cap:

Option	Trigger	Dispatch
1. Update our UPSTREAM (b/d, N total)	finding (b) > 0 OR finding (d) > 0	Skill(skill="/vp-beads:upstream-tracker", args=...) — args route to workflow 1 (Log a new entry) and/or workflow 7 (Sync from Basic Memory) inside upstream-tracker
2. File beads issues for sibling's friction (N)	finding (e) > 0	Bash → bd create per entry
3. Resolve cross-stale entries (N)	finding (g) > 0	Skill(skill="/vp-beads:upstream-tracker", args=...) → workflow 3 (Resolve an entry)
0. None — upstream report only	always	exit UPSTREAM tier without action

If neither tier has actionable findings for a sibling, skip the AskUserQuestion call entirely for that sibling. If only one tier has actionable findings, issue a single-question call (the SDK supports 1-4 questions per call).

Per-action argument templates

Pass natural-language prose in the Skill tool's args field. The delegated skill receives the prose as narrative context. Templates:

• SYNERGY 2 (Log unreciprocated sibling entries): Log unreciprocated entries from sibling <sibling-name>: <bullet list of titles + sections>. Invoke workflow 1 (Log a synergy entry) for each.
• UPSTREAM 1 (Update our UPSTREAM, b/d collapse): From sibling-sync findings against <sibling-name>: adopt complementary workarounds for <package, title> entries (sibling's workaround text: <quoted>); also scan sibling-only entries <package, titles>. Use workflow 1 (Log a new entry) and workflow 7 (Sync from Basic Memory) as appropriate.
• UPSTREAM 3 (Resolve cross-stale entries): Resolve entries in UPSTREAM-<sibling-name>.md: <titles>. Verify against the sibling's recent changelog/tags first. Invoke workflow 3 (Resolve an entry) for each.

For UPSTREAM 2 (bd create), construct each call with:

• --title="<entry title from sibling>"
• --type=task always. The sibling's UPSTREAM file body is plain prose and lacks the structured sections that bd's validation.on-create=error requires for --type=bug (## Steps to Reproduce, ## Acceptance Criteria) and --type=feature (## Acceptance Criteria). Filing as task always succeeds; the user can refile the resulting issue with bd update --type=bug after adding the required sections, or supersede with a properly structured bug-type issue. Note this in the post-batch report: "Filed N task-type beads issues from sibling friction; refile as bug/feature with required sections if needed."
• --description="<entry body verbatim>\n\nSibling <sibling-name> tracks this against us in their UPSTREAM-<this-name>.md. Source section: <Bugs | Feature Requests | Upstream Opportunities>."

Precedence with --auto-reciprocate

The --auto-reciprocate flag is the explicit non-interactive consent path. Its semantics relative to the new menu:

Invocation	Behavior
--auto-reciprocate flag set	Skip the action menu entirely. Run workflow 4 (Apply reciprocation batch) directly with its existing per-entry [y/n/skip-rest] confirmation gate.
No flag, user picks Q1 option 1 (Apply reciprocal gaps)	Enter workflow 4 (Apply reciprocation batch) with the same per-entry confirmation. The menu surfaces the same path interactively.
No flag, user picks any "None" option	No writes. Default read-only contract holds.
No flag, user picks any non-"None" option (other than Q1 option 1)	Dispatch per the table above. The delegated skill applies its own confirmation gate.

Idempotency and re-runs

Most actions self-resolve on the next sibling-sync run:

• Q1 option 1 closes finding (a) (the sibling now has the reciprocal entry).
• Q1 option 2 closes finding (b) (this side now has the entry; titles match).
• Q2 option 1 closes findings (b) and (d) on the UPSTREAM side.
• Q2 option 3 annotates _(Resolved ...)_ so workflow 3 (Sync sibling UPSTREAM)'s matching logic suppresses finding (g) thereafter.

Only Q2 option 2 (bd create for finding (e)) does not self-resolve: the sibling's UPSTREAM-<this-name>.md still carries the entry until the sibling annotates it as resolved on their side. Finding (e) will re-fire on subsequent sibling-sync runs until that happens. This is expected behavior — it reminds the user the friction is real and the local bd issue tracks our intent. No "skip-already-filed" cache is maintained.

Failure modes

• Delegated Skill call returns "skill not found" or errors. Fall back to printing the original copy-paste hint and continue to the next sibling. Never abort the whole run on a delegation failure.
• Subagent context. When sibling-sync runs as a Task subagent (e.g., inside a swarm-wave research agent), AskUserQuestion is explicitly unavailable per the Anthropic Agent SDK (https://code.claude.com/docs/en/agent-sdk/user-input — "Limitations: Subagents"), and Skill tool calls may silently no-op (undocumented behavior — subject to change). In subagent context the entire action menu cannot fire: skip the AskUserQuestion call, print the original copy-paste hints from workflows 2 (Sync sibling SYNERGY) and 3 (Sync sibling UPSTREAM) under the existing "→" arrow style, and let the parent agent decide whether to re-invoke /sibling-sync directly. No formal subagent probe is required — best-effort detection by attempting the AskUserQuestion call and falling back on the SDK error suffices.
• bd create failure (e.g., missing required --description field for bug type). Report the specific failure and continue to the next entry; don't abort the whole run.

Sprint Workflow Integration

/sibling-sync runs as an optional parallel diagnostic alongside /synergy-tracker's review and trend-review workflows. Recommended cadences:

• Before /synergy-tracker workflow 4 (Trend review (quarterly)) — every 4th sprint, run /sibling-sync first so the trend review has up-to- date drift findings to act on.
• Before /synergy-tracker workflow 2 (Review open synergies) — optional; surfaces drift the per-side review wouldn't catch.
• After significant sibling activity — when the user knows the sibling shipped a release or restructured a skill, run /sibling-sync to catch resulting drift early.

This skill is read-only in its default mode, so it's safe to run proactively without commitment to any follow-up action.

Guidelines

• Read-only by default. Default invocations only call workflows 1 (Discover sibling(s)), 2 (Sync sibling SYNERGY), and 3 (Sync sibling UPSTREAM), surfacing findings only. Edit and Write are in allowed-tools solely to support workflow 4 (Apply reciprocation batch). Never mutate without --auto-reciprocate (or equivalent explicit user intent).

• Per-entry confirmation under --auto-reciprocate. Even with the flag, every write requires explicit per-entry confirmation. Mirrors /upstream-tracker workflow 7 (Sync from Basic Memory)'s confirmation pattern.

• Skip inaccessible siblings, don't error. A missing local-path is informational, not fatal. Continue with what's available and report what was skipped so the user can correct via .claude/synergy-registry.local.json.

• Title-keyed comparison runs in two explicit passes. Entries on the two sides are written by different sessions and naturally drift in title formatting; deterministic matching catches the obvious wins, judgment ratifies the residual ambiguous cases. The two passes are separate so the deterministic rule stays testable and the judgment rule stays bounded.

  • Pass 1 — deterministic lead-clause match. For each title: lowercase, collapse whitespace runs to a single space, then take the lead clause = the substring before the first occurrence of :, —, --, or ( (whichever is earliest; if none of those appears, the lead clause is the full normalized title). Two entries pair in pass 1 iff their normalized lead clauses are byte-identical. Examples that should pair here:

    • wc -l portability guard ↔ wc -l portability guard (|| count=0 + tr -d ' ')
    • edit_note append-with-section gotcha: independently documented by both plugins ↔ edit_note append-with-section gotcha — independently documented
    • Frontmatter features ↔ Frontmatter features (skills, user-invocable, effort)

  • Pass 2 — judgment on residuals only. For entries that did NOT pair in pass 1, scan the still-unmatched residuals on the other side once for qualifier-phrase reorderings or token rearrangements that clearly describe the same idea. Pair only when the subjects are unambiguously the same. Pass 2 examples:

    • PreCompact hook retired in vp-knowledge v0.28.0 ↔ PreCompact hook retired in v0.28.0 (qualifier prepositional phrase)
    • Skill invocation layering: three levels vs two levels ↔ Skill invocation layering: two-level vs three-level (token reordering after the colon)

    Pass 2 may NEVER override or relax pass 1: do not unmatch a pair pass 1 produced, and do not collapse two pass-1-residual entries that have a shared prefix but materially different scopes (Hook validation vs Hook validation regression test → leave both as one-sided). Rationale for the cost asymmetry: in default read-only mode, a duplicate entry surviving on both sides outlives sprint cycles silently, while an over-merge surfaces immediately at workflow 4 (Apply reciprocation batch)'s per-entry confirmation gate where the user can reject. Under --auto-reciprocate this asymmetry inverts — a false-positive pass-2 match can suppress a reciprocal entry that should be written. Therefore: workflow 4 (Apply reciprocation batch) re-runs pass 1 only and treats pass 2 matches as advisory candidates that REQUIRE the user's per-entry confirmation to count as matches (mirrors the existing write-confirmation gate; the spec defaults to caution at the mutation boundary).

• Stale threshold is inline. 8 sprints (≈ two trend-review cycles) for SYNERGY Status: aligned rows; 3 months for UPSTREAM entries; 6 months for the workflow 3 (Sync sibling UPSTREAM) Mode B "shipped" look-back horizon (findings (g) and (h)). Canonical definitions for the 8-sprint and 3-month thresholds live in /synergy-tracker workflow 4 (Trend review (quarterly)) and /upstream-tracker workflow 4 (Trend review (quarterly)). The 6-month Mode B horizon is /sibling-sync's own choice — broader than the staleness flag because it requires cross-side evidence, not just age. When the canonical thresholds change, this skill must be updated to match — the validate-plugin convention check (vp-beads-9we) catches bare workflow refs but not threshold drift.

• Canonical project-name derivation. Workflow 3 (Sync sibling UPSTREAM) Mode B needs this project's own name to compute UPSTREAM-<this-name>.md at the sibling's root; workflow 4 (Apply reciprocation batch) needs it to name SYNERGY-<this-project>.md on the sibling. Derivation uses a four-tier precedence (sibling-registry back-pointer → plugin manifest → package manifest → directory basename), followed by normalization. Full algorithm, worked examples, and limitations: skills/synergy-tracker/references/project-name-derivation.md. The same algorithm computes <sibling-name> from this project's synergy-registry.json (tier 3 for the sibling subject). If derivation fails, see Error handling below.

• No new SYNERGY/UPSTREAM sections. /sibling-sync only writes entries into existing section schemas (## Shared Patterns, ## Divergences, …). It does not introduce new section types. Schema evolution is /synergy-tracker's job.

• Companion to /vendor-sync. vendor-sync handles upstream → project drift (subtree pulls, UPSTREAM auto-resolve). sibling-sync handles peer-to-peer drift along two axes: SYNERGY reciprocation/status divergence (workflow 2 (Sync sibling SYNERGY)), and UPSTREAM friction tracked across both sides (workflow 3 (Sync sibling UPSTREAM) — both shared-dependency Mode A and reciprocal-friction Mode B). Both skills default to reporting / read-only paths and gate mutations on explicit user intent.

• Project tempo classification. When a sibling has been dormant for more than 90 days (git -C <sibling-path> rev-list --count --since="90 days ago" HEAD returns 0), surface findings under that sibling with a "(dormant — drift expected)" note. Don't suppress findings — the user may still want to apply reciprocations to dormant siblings to keep them in lockstep — but contextualize them.

Error handling

• Registry not found — tell the user this project has no .claude/synergy-registry.json. Offer to redirect to /synergy-tracker workflow 1 (Log a synergy entry), which includes guided registry creation at step 1b. If the user names a sibling inline, follow step 1b's instructions verbatim from this conversation (Claude Code has no real cross-skill handoff, so this means executing step 1b's prose in-session by re-reading skills/synergy-tracker/SKILL.md), then resume this skill from workflow 1 (Discover sibling(s)) step 2. Otherwise stop, and direct the user to invoke /synergy-tracker first and re-run /sibling-sync afterwards.
• All siblings inaccessible — report which paths were tried and stop. Suggest .claude/synergy-registry.local.json for per-machine path overrides.
• Sibling SYNERGY file missing — treat as zero entries and proceed. The comparison will surface "Unreciprocated entries on sibling" findings if applicable.
• Sibling UPSTREAM file present but malformed — report the parse error with the file path and skip that file's findings. Continue with the rest of the sibling's UPSTREAM files.
• --auto-reciprocate with zero reciprocal gaps — report "no reciprocal gaps to apply" and exit cleanly without touching any file.
• Project-name not derivable — if both .claude-plugin/plugin.json is absent (or has no name) AND the project root directory basename is empty (e.g., the working directory is /), skip workflow 3 (Sync sibling UPSTREAM) Mode B for every sibling and report the limitation in the workflow 3 (Sync sibling UPSTREAM) output. Mode A still runs normally; SYNERGY workflow 2 (Sync sibling SYNERGY) is unaffected.