diff --git a/.devague/current_plan b/.devague/current_plan
index 0c0cfea..04631d4 100644
--- a/.devague/current_plan
+++ b/.devague/current_plan
@@ -1 +1 @@
-devague-ships-a-sharper-end-to-end-method-a-guided
+devague-multi-repo-efforts
diff --git a/.devague/frames/devague-multi-repo-efforts.json b/.devague/frames/devague-multi-repo-efforts.json
new file mode 100644
index 0000000..70e539b
--- /dev/null
+++ b/.devague/frames/devague-multi-repo-efforts.json
@@ -0,0 +1,268 @@
+{
+  "slug": "devague-multi-repo-efforts",
+  "title": "devague multi-repo efforts",
+  "schema_version": 1,
+  "status": "exported",
+  "created": "2026-07-06T21:47:44Z",
+  "updated": "2026-07-06T21:49:32Z",
+  "claims": [
+    {
+      "id": "c1",
+      "kind": "announcement",
+      "text": "devague supports multi-repo efforts: one effort \u2014 frame, plan, and workforce fan-out \u2014 can span multiple repositories instead of stopping at the edge of a single repo",
+      "origin": "user",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h1",
+          "text": "honest only if a plan whose tasks target >=2 repos converges and exports without hand-editing state JSON",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c2",
+      "kind": "audience",
+      "text": "the operator agent + user running efforts that span AgentCulture sibling repos (e.g. devague + guildmaster + eidetic-cli), and the workforce subagents briefed per task",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h7",
+          "text": "the multi-repo surface is exercised entirely through the existing operator skills (scope/think/spec-to-plan/assign-to-workforce) \u2014 no new privileged tooling beyond what the audience already runs",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c3",
+      "kind": "before_state",
+      "text": "an effort lives in one cwd-relative .devague/ per repo (store.py:1-3 'the frames live in the repo being specced'; plan_store.py:18) \u2014 a cross-repo effort must be hand-split into disconnected per-repo frames with no shared convergence gate",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h8",
+          "text": "verified against the shipped code at 0.15.0: store.py:15-18 and plan_store.py:18-19 are cwd-relative and no repo field exists in frame.py or plan.py",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c4",
+      "kind": "why_it_matters",
+      "text": "the last Colleague-assisted effort showed real efforts scale past one repo; devague's honesty gates should cover the whole effort, not stop at the repo edge",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h9",
+          "text": "the four pain points are the user's own report (2026-07-07): no shared convergence, task-to-repo mapping, fan-out mechanics, review across repos \u2014 not inferred",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c5",
+      "kind": "after_state",
+      "text": "one effort spans repos end-to-end: tasks name their target repo, the waves payload briefs the workforce per repo, and the human gates stay explicit across all touched repos",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h10",
+          "text": "honest only if every element is deliverable by the decided shape: hub-repo state (q2), repo-tagged tasks in one plan (q3), per-repo PRs with worktree-merge dependency ordering (q4)",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c6",
+      "kind": "requirement",
+      "text": "a plan task can name the repo it targets \u2014 devague/plan.py Task carries summary/acceptance/deps/covers today, no repo dimension; the workforce brief needs one",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h2",
+          "text": "existing single-repo frames/plans load unchanged: repo-less tasks stay valid (fail-closed schema bump, default = the effort's home repo)",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c7",
+      "kind": "requirement",
+      "text": "the waves payload carries each task's repo target so the operator creates the worktree in the right repo \u2014 cli/_commands/plan.py:326 emits bare task ids today; natural carrier is the #53 t9 enriched payload",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h3",
+          "text": "the repo field in `waves --json` is descriptive text the operator resolves \u2014 devague never validates it against the filesystem",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c8",
+      "kind": "requirement",
+      "text": "assign-to-workforce covers the multi-repo fan-out: per-repo worktrees, per-repo TDD gates, and the final-PR gate across every touched repo \u2014 SKILL.md today hard-assumes one repo (one `../worktrees` root, one main branch, one final PR)",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h4",
+          "text": "the TDD gate runs per repo: a task's tests pass before and after the merge in that task's repo; a green suite in one repo never vouches for another",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c9",
+      "kind": "boundary",
+      "text": "the CLI stays deterministic and non-orchestrating (issue 20): multi-repo lands as descriptive metadata only \u2014 the CLI never clones a repo, never reads or writes another repo's files, never manages worktrees or backends",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h5",
+          "text": "a boundary audit (grep/bandit, as in #53 t14) shows this effort added no git plumbing, repo I/O, or subprocess orchestration to the devague package",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c10",
+      "kind": "non_goal",
+      "text": "no change to dependency-wave computation \u2014 plan.py:172 dependency_waves layers tasks purely by deps and is already repo-agnostic",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c11",
+      "kind": "non_goal",
+      "text": "devague is not becoming a workspace manager or monorepo tool: no repo discovery, no checkout management, no cross-repo git plumbing in the CLI",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c12",
+      "kind": "assumption",
+      "text": "multi-repo lands on top of the committed #53 sharper-method plan (t1-t14, unimplemented) without forking it \u2014 per-item instructions (t2/t5) and the enriched waves payload (t9) are carriers for repo targets, not competitors",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c13",
+      "kind": "assumption",
+      "text": "the motivating Colleague effort is the eidetic 3-layer memory plan spanning eidetic-cli, devague skills, and culture agents (eidetic recall, record of 2026-06-20) \u2014 to be confirmed by the user",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c14",
+      "kind": "success_signal",
+      "text": "one real effort touching >=2 repos runs scope -> frame -> plan -> fan-out end-to-end on the shipped surface, with each task built in its target repo and every touched repo getting its reviewed PR",
+      "origin": "llm",
+      "status": "confirmed",
+      "honesty_conditions": [
+        {
+          "id": "h6",
+          "text": "the end-to-end run is real (>=2 actual repos, committed as a worked example or e2e evidence), not a single-repo run relabeled",
+          "status": "confirmed"
+        }
+      ],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c15",
+      "kind": "decision",
+      "text": "pain points (user, 2026-07-07, q1): the Colleague effort lacked shared convergence, task-to-repo mapping, fan-out mechanics, and cross-repo review \u2014 all four are the problem to solve",
+      "origin": "user",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c16",
+      "kind": "decision",
+      "text": "state home (q2): the effort's hub repo holds the single .devague/ frame+plan; sibling repos are referenced by name in tasks \u2014 the cwd-relative store is unchanged",
+      "origin": "user",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c17",
+      "kind": "decision",
+      "text": "granularity (q3): one frame + one plan per effort with repo-tagged tasks; one convergence gate and one dependency graph across repos \u2014 no meta-frame layer",
+      "origin": "user",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c18",
+      "kind": "decision",
+      "text": "human gate 3 (q4): one final PR per touched repo, reviewed together as one effort; a cross-repo dependent task starts when its dependency's worktree merge (main-agent TDD gate) lands, not its PR merge",
+      "origin": "user",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    },
+    {
+      "id": "c19",
+      "kind": "decision",
+      "text": "sequencing (q5): this effort exports its own spec+plan building on the #53 carriers (per-item instructions t2/t5, enriched waves t9), implemented after t1-t14 \u2014 the committed #53 plan is not reopened",
+      "origin": "user",
+      "status": "confirmed",
+      "honesty_conditions": [],
+      "hard_questions": [],
+      "links": []
+    }
+  ],
+  "open_vagueness": [
+    {
+      "id": "v1",
+      "text": "exact schema shape and naming of the repo target (task-level field vs plan-level repo list vs instruction-carried) \u2014 a build-plan detail deferred to the /spec-to-plan leg",
+      "kind": "unknown_nonblocking",
+      "claim_id": null
+    }
+  ]
+}
diff --git a/.devague/plans/devague-multi-repo-efforts.json b/.devague/plans/devague-multi-repo-efforts.json
new file mode 100644
index 0000000..5f8f136
--- /dev/null
+++ b/.devague/plans/devague-multi-repo-efforts.json
@@ -0,0 +1,280 @@
+{
+  "slug": "devague-multi-repo-efforts",
+  "title": "devague multi-repo efforts",
+  "frame_slug": "devague-multi-repo-efforts",
+  "schema_version": 1,
+  "status": "exported",
+  "created": "2026-07-06T21:52:13Z",
+  "updated": "2026-07-06T21:55:36Z",
+  "targets": [
+    {
+      "id": "c1",
+      "kind": "announcement",
+      "text": "devague supports multi-repo efforts: one effort \u2014 frame, plan, and workforce fan-out \u2014 can span multiple repositories instead of stopping at the edge of a single repo"
+    },
+    {
+      "id": "h1",
+      "kind": "honesty",
+      "text": "honest only if a plan whose tasks target >=2 repos converges and exports without hand-editing state JSON"
+    },
+    {
+      "id": "c2",
+      "kind": "audience",
+      "text": "the operator agent + user running efforts that span AgentCulture sibling repos (e.g. devague + guildmaster + eidetic-cli), and the workforce subagents briefed per task"
+    },
+    {
+      "id": "h7",
+      "kind": "honesty",
+      "text": "the multi-repo surface is exercised entirely through the existing operator skills (scope/think/spec-to-plan/assign-to-workforce) \u2014 no new privileged tooling beyond what the audience already runs"
+    },
+    {
+      "id": "c3",
+      "kind": "before_state",
+      "text": "an effort lives in one cwd-relative .devague/ per repo (store.py:1-3 'the frames live in the repo being specced'; plan_store.py:18) \u2014 a cross-repo effort must be hand-split into disconnected per-repo frames with no shared convergence gate"
+    },
+    {
+      "id": "h8",
+      "kind": "honesty",
+      "text": "verified against the shipped code at 0.15.0: store.py:15-18 and plan_store.py:18-19 are cwd-relative and no repo field exists in frame.py or plan.py"
+    },
+    {
+      "id": "c4",
+      "kind": "why_it_matters",
+      "text": "the last Colleague-assisted effort showed real efforts scale past one repo; devague's honesty gates should cover the whole effort, not stop at the repo edge"
+    },
+    {
+      "id": "h9",
+      "kind": "honesty",
+      "text": "the four pain points are the user's own report (2026-07-07): no shared convergence, task-to-repo mapping, fan-out mechanics, review across repos \u2014 not inferred"
+    },
+    {
+      "id": "c5",
+      "kind": "after_state",
+      "text": "one effort spans repos end-to-end: tasks name their target repo, the waves payload briefs the workforce per repo, and the human gates stay explicit across all touched repos"
+    },
+    {
+      "id": "h10",
+      "kind": "honesty",
+      "text": "honest only if every element is deliverable by the decided shape: hub-repo state (q2), repo-tagged tasks in one plan (q3), per-repo PRs with worktree-merge dependency ordering (q4)"
+    },
+    {
+      "id": "c6",
+      "kind": "requirement",
+      "text": "a plan task can name the repo it targets \u2014 devague/plan.py Task carries summary/acceptance/deps/covers today, no repo dimension; the workforce brief needs one"
+    },
+    {
+      "id": "h2",
+      "kind": "honesty",
+      "text": "existing single-repo frames/plans load unchanged: repo-less tasks stay valid (fail-closed schema bump, default = the effort's home repo)"
+    },
+    {
+      "id": "c7",
+      "kind": "requirement",
+      "text": "the waves payload carries each task's repo target so the operator creates the worktree in the right repo \u2014 cli/_commands/plan.py:326 emits bare task ids today; natural carrier is the #53 t9 enriched payload"
+    },
+    {
+      "id": "h3",
+      "kind": "honesty",
+      "text": "the repo field in `waves --json` is descriptive text the operator resolves \u2014 devague never validates it against the filesystem"
+    },
+    {
+      "id": "c8",
+      "kind": "requirement",
+      "text": "assign-to-workforce covers the multi-repo fan-out: per-repo worktrees, per-repo TDD gates, and the final-PR gate across every touched repo \u2014 SKILL.md today hard-assumes one repo (one `../worktrees` root, one main branch, one final PR)"
+    },
+    {
+      "id": "h4",
+      "kind": "honesty",
+      "text": "the TDD gate runs per repo: a task's tests pass before and after the merge in that task's repo; a green suite in one repo never vouches for another"
+    },
+    {
+      "id": "c9",
+      "kind": "boundary",
+      "text": "the CLI stays deterministic and non-orchestrating (issue 20): multi-repo lands as descriptive metadata only \u2014 the CLI never clones a repo, never reads or writes another repo's files, never manages worktrees or backends"
+    },
+    {
+      "id": "h5",
+      "kind": "honesty",
+      "text": "a boundary audit (grep/bandit, as in #53 t14) shows this effort added no git plumbing, repo I/O, or subprocess orchestration to the devague package"
+    },
+    {
+      "id": "c14",
+      "kind": "success_signal",
+      "text": "one real effort touching >=2 repos runs scope -> frame -> plan -> fan-out end-to-end on the shipped surface, with each task built in its target repo and every touched repo getting its reviewed PR"
+    },
+    {
+      "id": "h6",
+      "kind": "honesty",
+      "text": "the end-to-end run is real (>=2 actual repos, committed as a worked example or e2e evidence), not a single-repo run relabeled"
+    }
+  ],
+  "tasks": [
+    {
+      "id": "t1",
+      "summary": "Plan schema: optional repo target on Task (plan.py, plan_store.py)",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "a task with a repo target round-trips save/load verbatim; the field is descriptive text (a repo name), owned by plan.py + plan_store.py only",
+        "pre-existing plans and repo-less tasks load with no error \u2014 an absent repo means the effort's home repo (h2); plan schema_version bumped once, fail-closed, on top of the #53 t2 bump"
+      ],
+      "deps": [],
+      "covers": [
+        "c6",
+        "h2"
+      ]
+    },
+    {
+      "id": "t2",
+      "summary": "CLI: `plan task --repo <name>` flag + `retarget <tN> --repo <name>` move (cli/_commands/plan.py)",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "`plan task --repo <name>` stores the repo target verbatim and `plan show --json` includes it per task",
+        "a new `retarget <tN> --repo <name>` move sets or updates the target on an existing task and flips a confirmed task back to proposed \u2014 the user re-confirms (mirrors the #53 instruction re-confirm rule)"
+      ],
+      "deps": [
+        "t1"
+      ],
+      "covers": [
+        "c6"
+      ]
+    },
+    {
+      "id": "t3",
+      "summary": "Waves: repo target in the enriched `waves --json` payload (cli/_commands/plan.py)",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "every task entry in `waves --json` carries its repo target verbatim as descriptive text the operator resolves \u2014 no filesystem validation anywhere (h3); repo-less tasks emit the home-repo default",
+        "output stays deterministic and read-only; the dependency-wave computation itself is untouched (frame non-goal c10)"
+      ],
+      "deps": [
+        "t1",
+        "t2"
+      ],
+      "covers": [
+        "c7",
+        "h3"
+      ]
+    },
+    {
+      "id": "t4",
+      "summary": "Docs: spec-contract + hub-repo state model (docs/spec-contract.md, docs/llm-guidance.md, docs/skills.md)",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "docs/spec-contract.md documents the task repo target, the hub-repo state home (the cwd-relative store \u2014 store.py:15-18, plan_store.py:18-19 \u2014 stays unchanged, cited), and the plan schema bump",
+        "docs/llm-guidance.md and docs/skills.md updated to match; the single-repo default is described as the unchanged base case (h8)"
+      ],
+      "deps": [
+        "t1"
+      ],
+      "covers": [
+        "c3",
+        "h8"
+      ]
+    },
+    {
+      "id": "t5",
+      "summary": "assign-to-workforce: multi-repo fan-out method (SKILL.md)",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "SKILL.md documents per-repo worktrees (one per task, created in that task's target repo), the per-repo TDD gate \u2014 a task's tests pass before and after the merge in that task's repo; a green suite in one repo never vouches for another (h4)",
+        "gate 3 documented as one final PR per touched repo, reviewed together as one effort; cross-repo deps are satisfied at worktree-merge time, not PR-merge time (decision c18); the single-repo flow reads unchanged"
+      ],
+      "deps": [
+        "t3"
+      ],
+      "covers": [
+        "c8",
+        "h4"
+      ]
+    },
+    {
+      "id": "t6",
+      "summary": "split-plan script: repo column from the enriched payload (assign-to-workforce.sh)",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "split-plan renders each task's repo target from `waves --json` in the implementation split table; single-repo plans render unchanged",
+        "the script stays skill-side \u2014 no CLI change, no orchestration added to the devague package"
+      ],
+      "deps": [
+        "t3",
+        "t5"
+      ],
+      "covers": [
+        "c8",
+        "c2"
+      ]
+    },
+    {
+      "id": "t7",
+      "summary": "Operator skills + CLI teaching surface (/spec-to-plan, /think, /scope SKILL.md; learn.py, explain.py)",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "/spec-to-plan documents `--repo <name>` and `retarget <tN>` with a worked multi-repo example; /think and /scope document the hub-repo state home for a multi-repo effort",
+        "`devague plan learn` and `devague plan explain` cover the repo target; the whole surface is exercised through the four existing operator skills \u2014 no new privileged tooling (h7)"
+      ],
+      "deps": [
+        "t2",
+        "t3"
+      ],
+      "covers": [
+        "c2",
+        "h7"
+      ]
+    },
+    {
+      "id": "t8",
+      "summary": "E2E dogfood across >=2 real repos + boundary audit",
+      "origin": "llm",
+      "status": "confirmed",
+      "acceptance_criteria": [
+        "one real effort touching >=2 actual repos (hub: this repo; candidate sibling: guildmaster or eidetic-cli) runs scope -> frame -> plan -> fan-out on the shipped surface, committed as a worked example or e2e evidence (h6), converging and exporting without hand-editing state JSON (h1)",
+        "boundary audit (grep/bandit, the #53 t14 pattern) shows no git plumbing, repo I/O, or subprocess orchestration added to the devague package (c9/h5)",
+        "the worked example demonstrably addresses the four user-reported pain points \u2014 shared convergence, task-to-repo mapping, fan-out mechanics, cross-repo review (c4/h9) \u2014 in the decided shape: hub-repo state, repo-tagged tasks, per-repo PRs (c5/h10)"
+      ],
+      "deps": [
+        "t4",
+        "t6",
+        "t7"
+      ],
+      "covers": [
+        "c1",
+        "h1",
+        "c4",
+        "h9",
+        "c5",
+        "h10",
+        "c9",
+        "h5",
+        "c14",
+        "h6"
+      ]
+    }
+  ],
+  "risks": [
+    {
+      "id": "r1",
+      "text": "external precondition: the #53 build (t2/t5/t9 there \u2014 instruction fields + enriched waves payload) lands before this plan's t1/t3; a dependency on another plan's tasks cannot be recorded in this graph (decision c19)",
+      "kind": "follow_up",
+      "task_id": "t1"
+    },
+    {
+      "id": "r2",
+      "text": "the e2e run needs a second real repo willing to take a dogfood PR \u2014 candidate guildmaster or eidetic-cli; pick at execution time",
+      "kind": "unknown_nonblocking",
+      "task_id": "t8"
+    },
+    {
+      "id": "r3",
+      "text": "`plan confirm` takes one id per call (no transactional multi-id parity with the frame engine yet \u2014 recorded follow-up in #53); batch-confirming multi-repo plans stays manual until then",
+      "kind": "follow_up",
+      "task_id": null
+    }
+  ]
+}
diff --git a/CHANGELOG.md b/CHANGELOG.md
index bfaed7c..19ef0ab 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,23 @@ All notable changes to this project will be documented in this file.
 Format follows [Keep a Changelog](https://keepachangelog.com/). This project
 adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.15.1] - 2026-07-07
+
+### Added
+
+- Exported **spec and plan artifacts** for the "multi-repo efforts" increment
+  (docs/specs + docs/plans 2026-07-06, plus the frame/plan state JSON as the
+  evidence trail; produced by a dogfooded /scope → /think → /spec-to-plan run).
+  **Documentation and state only — no CLI changes ship in this release.** The
+  decided shape: hub-repo state home (the cwd-relative store is unchanged), one
+  plan per effort with repo-tagged tasks (single gate, single dependency
+  graph), the repo target as descriptive metadata riding the #53 carriers
+  (per-item instructions, enriched waves payload), per-repo worktrees + TDD
+  gates in assign-to-workforce, and one final PR per touched repo with
+  cross-repo dependencies satisfied at worktree-merge time. Implementation is
+  the committed plan (8 tasks over 6 file-disjoint waves), sequenced **after**
+  the #53 build (t1–t14) and explicitly not reopening it.
+
 ## [0.14.1] - 2026-07-02
 
 ### Added
diff --git a/docs/plans/2026-07-06-devague-multi-repo-efforts.md b/docs/plans/2026-07-06-devague-multi-repo-efforts.md
new file mode 100644
index 0000000..ad991cf
--- /dev/null
+++ b/docs/plans/2026-07-06-devague-multi-repo-efforts.md
@@ -0,0 +1,77 @@
+# Build Plan — devague multi-repo efforts
+
+slug: `devague-multi-repo-efforts` · status: `exported` · from frame: `devague-multi-repo-efforts`
+
+> devague supports multi-repo efforts: one effort — frame, plan, and workforce fan-out — can span multiple repositories instead of stopping at the edge of a single repo
+
+## Tasks
+
+### t1 — Plan schema: optional repo target on Task (plan.py, plan_store.py)
+
+- covers: c6, h2
+- acceptance:
+  - a task with a repo target round-trips save/load verbatim; the field is descriptive text (a repo name), owned by plan.py + plan_store.py only
+  - pre-existing plans and repo-less tasks load with no error — an absent repo means the effort's home repo (h2); plan schema_version bumped once, fail-closed, on top of the #53 t2 bump
+
+### t2 — CLI: `plan task --repo <name>` flag + `retarget <tN> --repo <name>` move (cli/_commands/plan.py)
+
+- depends on: t1
+- covers: c6
+- acceptance:
+  - `plan task --repo <name>` stores the repo target verbatim and `plan show --json` includes it per task
+  - a new `retarget <tN> --repo <name>` move sets or updates the target on an existing task and flips a confirmed task back to proposed — the user re-confirms (mirrors the #53 instruction re-confirm rule)
+
+### t3 — Waves: repo target in the enriched `waves --json` payload (cli/_commands/plan.py)
+
+- depends on: t1, t2
+- covers: c7, h3
+- acceptance:
+  - every task entry in `waves --json` carries its repo target verbatim as descriptive text the operator resolves — no filesystem validation anywhere (h3); repo-less tasks emit the home-repo default
+  - output stays deterministic and read-only; the dependency-wave computation itself is untouched (frame non-goal c10)
+
+### t4 — Docs: spec-contract + hub-repo state model (docs/spec-contract.md, docs/llm-guidance.md, docs/skills.md)
+
+- depends on: t1
+- covers: c3, h8
+- acceptance:
+  - docs/spec-contract.md documents the task repo target, the hub-repo state home (the cwd-relative store — store.py:15-18, plan_store.py:18-19 — stays unchanged, cited), and the plan schema bump
+  - docs/llm-guidance.md and docs/skills.md updated to match; the single-repo default is described as the unchanged base case (h8)
+
+### t5 — assign-to-workforce: multi-repo fan-out method (SKILL.md)
+
+- depends on: t3
+- covers: c8, h4
+- acceptance:
+  - SKILL.md documents per-repo worktrees (one per task, created in that task's target repo), the per-repo TDD gate — a task's tests pass before and after the merge in that task's repo; a green suite in one repo never vouches for another (h4)
+  - gate 3 documented as one final PR per touched repo, reviewed together as one effort; cross-repo deps are satisfied at worktree-merge time, not PR-merge time (decision c18); the single-repo flow reads unchanged
+
+### t6 — split-plan script: repo column from the enriched payload (assign-to-workforce.sh)
+
+- depends on: t3, t5
+- covers: c8, c2
+- acceptance:
+  - split-plan renders each task's repo target from `waves --json` in the implementation split table; single-repo plans render unchanged
+  - the script stays skill-side — no CLI change, no orchestration added to the devague package
+
+### t7 — Operator skills + CLI teaching surface (/spec-to-plan, /think, /scope SKILL.md; learn.py, explain.py)
+
+- depends on: t2, t3
+- covers: c2, h7
+- acceptance:
+  - /spec-to-plan documents `--repo <name>` and `retarget <tN>` with a worked multi-repo example; /think and /scope document the hub-repo state home for a multi-repo effort
+  - `devague plan learn` and `devague plan explain` cover the repo target; the whole surface is exercised through the four existing operator skills — no new privileged tooling (h7)
+
+### t8 — E2E dogfood across >=2 real repos + boundary audit
+
+- depends on: t4, t6, t7
+- covers: c1, h1, c4, h9, c5, h10, c9, h5, c14, h6
+- acceptance:
+  - one real effort touching >=2 actual repos (hub: this repo; candidate sibling: guildmaster or eidetic-cli) runs scope -> frame -> plan -> fan-out on the shipped surface, committed as a worked example or e2e evidence (h6), converging and exporting without hand-editing state JSON (h1)
+  - boundary audit (grep/bandit, the #53 t14 pattern) shows no git plumbing, repo I/O, or subprocess orchestration added to the devague package (c9/h5)
+  - the worked example demonstrably addresses the four user-reported pain points — shared convergence, task-to-repo mapping, fan-out mechanics, cross-repo review (c4/h9) — in the decided shape: hub-repo state, repo-tagged tasks, per-repo PRs (c5/h10)
+
+## Risks
+
+- [follow_up] external precondition: the #53 build (t2/t5/t9 there — instruction fields + enriched waves payload) lands before this plan's t1/t3; a dependency on another plan's tasks cannot be recorded in this graph (decision c19) (task t1)
+- [unknown_nonblocking] the e2e run needs a second real repo willing to take a dogfood PR — candidate guildmaster or eidetic-cli; pick at execution time (task t8)
+- [follow_up] `plan confirm` takes one id per call (no transactional multi-id parity with the frame engine yet — recorded follow-up in #53); batch-confirming multi-repo plans stays manual until then
diff --git a/docs/specs/2026-07-06-devague-multi-repo-efforts.md b/docs/specs/2026-07-06-devague-multi-repo-efforts.md
new file mode 100644
index 0000000..c53bbff
--- /dev/null
+++ b/docs/specs/2026-07-06-devague-multi-repo-efforts.md
@@ -0,0 +1,61 @@
+# devague multi-repo efforts
+
+> devague supports multi-repo efforts: one effort — frame, plan, and workforce fan-out — can span multiple repositories instead of stopping at the edge of a single repo
+
+## Audience
+
+- the operator agent + user running efforts that span AgentCulture sibling repos (e.g. devague + guildmaster + eidetic-cli), and the workforce subagents briefed per task
+
+## Before → After
+
+- Before: an effort lives in one cwd-relative .devague/ per repo (store.py:1-3 'the frames live in the repo being specced'; plan_store.py:18) — a cross-repo effort must be hand-split into disconnected per-repo frames with no shared convergence gate
+- After: one effort spans repos end-to-end: tasks name their target repo, the waves payload briefs the workforce per repo, and the human gates stay explicit across all touched repos
+
+## Why it matters
+
+- the last Colleague-assisted effort showed real efforts scale past one repo; devague's honesty gates should cover the whole effort, not stop at the repo edge
+
+## Requirements
+
+- a plan task can name the repo it targets — devague/plan.py Task carries summary/acceptance/deps/covers today, no repo dimension; the workforce brief needs one
+  - honesty: existing single-repo frames/plans load unchanged: repo-less tasks stay valid (fail-closed schema bump, default = the effort's home repo)
+- the waves payload carries each task's repo target so the operator creates the worktree in the right repo — cli/_commands/plan.py:326 emits bare task ids today; natural carrier is the #53 t9 enriched payload
+  - honesty: the repo field in `waves --json` is descriptive text the operator resolves — devague never validates it against the filesystem
+- assign-to-workforce covers the multi-repo fan-out: per-repo worktrees, per-repo TDD gates, and the final-PR gate across every touched repo — SKILL.md today hard-assumes one repo (one `../worktrees` root, one main branch, one final PR)
+  - honesty: the TDD gate runs per repo: a task's tests pass before and after the merge in that task's repo; a green suite in one repo never vouches for another
+
+## Honesty conditions
+
+- honest only if a plan whose tasks target >=2 repos converges and exports without hand-editing state JSON
+- the multi-repo surface is exercised entirely through the existing operator skills (scope/think/spec-to-plan/assign-to-workforce) — no new privileged tooling beyond what the audience already runs
+- verified against the shipped code at 0.15.0: store.py:15-18 and plan_store.py:18-19 are cwd-relative and no repo field exists in frame.py or plan.py
+- the four pain points are the user's own report (2026-07-07): no shared convergence, task-to-repo mapping, fan-out mechanics, review across repos — not inferred
+- honest only if every element is deliverable by the decided shape: hub-repo state (q2), repo-tagged tasks in one plan (q3), per-repo PRs with worktree-merge dependency ordering (q4)
+- a boundary audit (grep/bandit, as in #53 t14) shows this effort added no git plumbing, repo I/O, or subprocess orchestration to the devague package
+- the end-to-end run is real (>=2 actual repos, committed as a worked example or e2e evidence), not a single-repo run relabeled
+
+## Success signals
+
+- one real effort touching >=2 repos runs scope -> frame -> plan -> fan-out end-to-end on the shipped surface, with each task built in its target repo and every touched repo getting its reviewed PR
+
+## Scope / boundaries
+
+- the CLI stays deterministic and non-orchestrating (issue 20): multi-repo lands as descriptive metadata only — the CLI never clones a repo, never reads or writes another repo's files, never manages worktrees or backends
+
+## Non-goals
+
+- no change to dependency-wave computation — plan.py:172 dependency_waves layers tasks purely by deps and is already repo-agnostic
+- devague is not becoming a workspace manager or monorepo tool: no repo discovery, no checkout management, no cross-repo git plumbing in the CLI
+
+## Assumptions
+
+- multi-repo lands on top of the committed #53 sharper-method plan (t1-t14, unimplemented) without forking it — per-item instructions (t2/t5) and the enriched waves payload (t9) are carriers for repo targets, not competitors
+- the motivating Colleague effort is the eidetic 3-layer memory plan spanning eidetic-cli, devague skills, and culture agents (eidetic recall, record of 2026-06-20) — to be confirmed by the user
+
+## Decisions
+
+- pain points (user, 2026-07-07, q1): the Colleague effort lacked shared convergence, task-to-repo mapping, fan-out mechanics, and cross-repo review — all four are the problem to solve
+- state home (q2): the effort's hub repo holds the single .devague/ frame+plan; sibling repos are referenced by name in tasks — the cwd-relative store is unchanged
+- granularity (q3): one frame + one plan per effort with repo-tagged tasks; one convergence gate and one dependency graph across repos — no meta-frame layer
+- human gate 3 (q4): one final PR per touched repo, reviewed together as one effort; a cross-repo dependent task starts when its dependency's worktree merge (main-agent TDD gate) lands, not its PR merge
+- sequencing (q5): this effort exports its own spec+plan building on the #53 carriers (per-item instructions t2/t5, enriched waves t9), implemented after t1-t14 — the committed #53 plan is not reopened
diff --git a/pyproject.toml b/pyproject.toml
index e15dcbd..e3c49e1 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "devague"
-version = "0.14.1"
+version = "0.15.1"
 description = "devague — turns a vague feature idea into a buildable spec, then a buildable plan."
 readme = "README.md"
 license = "MIT"
diff --git a/uv.lock b/uv.lock
index 54150ce..e689896 100644
--- a/uv.lock
+++ b/uv.lock
@@ -183,7 +183,7 @@ wheels = [
 
 [[package]]
 name = "devague"
-version = "0.13.0"
+version = "0.15.1"
 source = { editable = "." }
 
 [package.dev-dependencies]