There's a project named `tangols` that I have configured to ingest up

You are being dispatched from flower Brief #210: There's a project named `tangols` that I have configured to ingest up

Recall pointer:
- Use recall_brief with id 210 for the full folder if you need provenance.

Target:
- project: flower (/Users/mikeferrara/Documents/code/flower)
- branch: flower/210-relocated-repo-session-correlation
- worktree: not specified
- kind: fresh

Current brief spec:
## Problem
flower attributes ingested **sessions → projects by cwd path only**. `SessionRegistryResolver::resolveProjectAndWorktree(?string $cwd)` matches a session's cwd against each `Project->root_path` (longest-prefix wins; `worktree.path` checked first). When a repo's checkout **moves** (e.g. tangols: `~/Herd/lsnet` → `~/Documents/code/tangols`, 2026-06-27), sessions recorded under the old path no longer match the relocated project's `root_path`, so they don't surface under it. Commits already correlate **per-repo** (P7 `flower:ingest-commits` via `GitInspector`), so this gap is session-specific.

## Decision (operator, 2026-07-04)
- **Scope (Q45):** build the **general relocated-repo correlation feature** (fix the whole class) — not a one-off re-map.
- **Strategy (Q46):** correlate a session to its project by **git repo identity** (remote URL → `owner/repo`), mirroring commit ingest.

## Approach
1. **Repo-identity primitive** — reuse `GitInspector`: `remoteUrl($dir)` + static `ownerRepo($url)` → canonical `owner/repo` (host-qualified where possible to avoid fork collisions). Fallback for remote-less repos: root-commit hash (`git rev-list --max-parents=0 HEAD`). Store the full remote URL as the key; match on `owner/repo`.
2. **Capture at INGEST (critical):** derive + **persist repo identity on the session row at ingest time**, while the cwd still exists on disk (old paths vanish after a move — see Constraint). Touch points: `app/Jobs/IngestSession.php` + the harness adapters (`app/Harness/Adapters/*Adapter.php`) that build the NormalizedSession.
3. **Store project identity:** persist each Project's repo identity (new column, e.g. `projects.repo_identity`, computed from its `root_path` git remote) so resolution is a cheap indexed lookup, not a per-session shell-out.
4. **Resolver fallback** (`SessionRegistryResolver`): keep worktree-path + project-path matching as the **primary/fast path** (correct when the checkout hasn't moved; preserves worktree precedence, no regression). When path match fails, fall back to **repo-identity match** (`session.repo_identity == project.repo_identity`) so relocated sessions attribute to the current project.
5. **Backfill / re-attribution:** a command (or extend `ReconcileIngestState`) that re-resolves already-ingested sessions and re-points those whose attribution changes under repo-identity. Best-effort: use stored session repo_identity if present; else derive from the on-disk repo when the old path still exists; else skip.

## Key constraint / honest edge (validated 2026-07-04)
The motivating session is the **hard case**: `~/Herd/lsnet` is **already gone from disk** (verified — relocated to `~/Documents/code/tangols`, remote `git@github.com:mferrara/tangols.git` = `mferrara/tangols`). Its repo identity cannot be derived by shelling the vanished path now. Recoverable only if (a) flower captured identity at original ingest, or (b) the CC `.jsonl` carries git-remote metadata (worker: check whether `ClaudeAdapter` exposes a remote — the resolver currently takes only cwd). Otherwise this 1 historical session may remain unattributed — **acceptable**, since the operator confirmed tangols's "none" is mostly genuine sparsity. The feature's real value is all **future + still-on-disk** relocations.

## Edge cases
- Distinct repos, same basename (e.g. `./lsnet` ≠ `./tangols` per code catalog) → key on **remote URL/host**, never directory basename.
- Repos with no remote → root-commit-hash identity, else stay path-based.
- Worktrees / multiple checkouts of one repo → `worktree.path` match keeps precedence; repo-identity is last-resort only.
- cwd in a subdir of the repo → resolve to git toplevel before deriving identity.
- Non-git cwds → unchanged (path-based).

## Acceptance
- New sessions under any cwd of a project's repo attribute to that project even after the checkout moves.
- Historical sessions whose **old path still exists** re-attribute on backfill.
- Path-based resolution unchanged when valid (no regression; worktree precedence intact); no cross-repo mis-attribution (distinct remotes stay distinct).
- tangols: documented best-effort — attributes if identity recoverable from stored/session data; else acknowledged as genuinely sparse.

## Prior art / touch points
`SessionRegistryResolver` (resolver), `GitInspector::remoteUrl` / `::ownerRepo` (identity, reference impl), `IngestSession` + harness adapters (capture), `CommitIngestService` / `IngestCommits` (repo-keyed reference), `ReconcileIngestState` (backfill host). Related: todo #340 (ingest core), scratchpad 1007 (umbrella multi-project SPEC), brief #153 (cwd mis-attribution, complete). Cross-link (NOT a dependency): **#196** cross-project support (PARKED) — this is the distinct "attribution across a path move" slice.

_Status → planned. Refine stops here; orchestrator owns dispatch._

Recent/key trace events:
[1] participant_joined operator:mike: (no body)
[2] note_added operator:mike: There's a project named `tangols` that I have configured to ingest up to 360 days worth of sessions/etc and it says there are none. I suspect that this is due to this repo potentially being moved from ~/Herd or somewhere in that ~/Herd tree and into where it is now and our session parsing effectively not picking up that move. Is there any way in our parsing/ingesting that we could correlate the sessions? (assuming they're there under a different path - maybe this would/should already work based on the repo and not the path and there really just are no sessions)
[3] status_change operator:mike: (no body)
[4] participant_joined flower-refine: (no body)
[5] refinement flower-refine: **Refinement (flower-refine gen-5, 2026-07-04) — grounded diagnosis + evidence**

**Confirmed mechanism:** session→project attribution is **cwd-based**, not repo-based.
- P2 Ingest-core (todo #340): the Claude adapter parses `~/.claude/projects/**/*.jsonl` and *"resolve[s] project/worktree by cwd"*.
- SPEC scratchpad 1007 (umbrella multi-project): *"Sessions are attributed to a project by cwd."*
- Brief #153 (complete): worktree sessions mis-attributed to `code` by cwd — same root cause family.
- Contrast: commits already ingest **per-repo** (P7 `flower:ingest-commits`), so a repo-identity correlation path already exists on the commit side but NOT for sessions.

**Diagnostic run (Claude store only, read-only):**
- tangols relocated `~/Herd/lsnet` → `~/Documents/code/tangols` on 2026-06-27 (code catalog).
- `~/.claude/projects/-Users-mikeferrara-Herd-lsnet/`: **1** session, newest mtime **2026-06-27** (≈ the move date).
- `~/.claude/projects/-Users-mikeferrara-Documents-code-tangols/`: **does not exist** → 0 CC sessions under the new path since the move.
- NOT checked: Codex (`~/.codex`) / Pi stores — the "360-day" config may also expect those.

**Interpretation:** tangols showing "none" is **mostly legitimate sparsity** (≈no CC activity under the new path since the move) **plus** a genuine but tiny path-correlation gap: the 1 old-path session won't attribute to the relocated tangols project under cwd resolution. So #210 is a **low-volume but clean instance of the general class** "cwd attribution breaks when a repo moves." The operator's own note anticipated this ("maybe there really just are no sessions") — and it's ~correct.

**Relationship to #196 (parked):** distinct. #196's two halves are (A) cross-project recall *grouping* and (B) conductor job-routing. #210 is a third thing — single-project attribution **across a path move**. Recommend **stand-alone brief, cross-linked to #196**, not folded in.

**Candidate approaches (for the general fix):**
(A) Correlate sessions by **repo identity** (git remote URL and/or root-commit hash) → any cwd of a repo maps to one project; mirrors commit ingest.
(B) Per-project **path-history / aliases** (record prior root paths on move) → re-attribute old-path sessions to the current project.
(C) One-off **re-map / backfill** tool (operator points an old path → project).

Two decisions opened for the operator (scope, strategy). Holding at `refining` pending answers — not proposing a canonical spec until scope is set.
[6] agent_question flower-refine: Diagnostic found only ~1 stranded tangols session in the Claude store (newest 2026-06-27, the move date) and zero under the new path — so 'none' is mostly genuine sparsity plus a tiny path-correlation gap. What outcome do you want?
[7] agent_question flower-refine: If we build the general fix, how should a session correlate to its project across a repo move?
[8] operator_answer operator:mike: {"choice":"By git repo identity (remote URL and/or root-commit hash) — mirrors commit ingest"}
[9] operator_answer operator:mike: {"choice":"General relocated-repo correlation feature (fix the whole class)"}
[10] spec_snapshot flower-refine: There's a project named `tangols` that I have configured to ingest up to 360 days worth of sessions/etc and it says there are none. I suspect that this is due to this repo potentially being moved from ~/Herd or somewhere in that ~/Herd tree and into where it is now and our session parsing effectively not picking up that move. Is there any way in our parsing/ingesting that we could correlate the sessions? (assuming they're there under a different path - maybe this would/should already work based on the repo and not the path and there really just are no sessions)
[11] refinement flower-refine: ## Problem
flower attributes ingested **sessions → projects by cwd path only**. `SessionRegistryResolver::resolveProjectAndWorktree(?string $cwd)` matches a session's cwd against each `Project->root_path` (longest-prefix wins; `worktree.path` checked first). When a repo's checkout **moves** (e.g. tangols: `~/Herd/lsnet` → `~/Documents/code/tangols`, 2026-06-27), sessions recorded under the old path no longer match the relocated project's `root_path`, so they don't surface under it. Commits already correlate **per-repo** (P7 `flower:ingest-commits` via `GitInspector`), so this gap is session-specific.

## Decision (operator, 2026-07-04)
- **Scope (Q45):** build the **general relocated-repo correlation feature** (fix the whole class) — not a one-off re-map.
- **Strategy (Q46):** correlate a session to its project by **git repo identity** (remote URL → `owner/repo`), mirroring commit ingest.

## Approach
1. **Repo-identity primitive** — reuse `GitInspector`: `remoteUrl($dir)` + static `ownerRepo($url)` → canonical `owner/repo` (host-qualified where possible to avoid fork collisions). Fallback for remote-less repos: root-commit hash (`git rev-list --max-parents=0 HEAD`). Store the full remote URL as the key; match on `owner/repo`.
2. **Capture at INGEST (critical):** derive + **persist repo identity on the session row at ingest time**, while the cwd still exists on disk (old paths vanish after a move — see Constraint). Touch points: `app/Jobs/IngestSession.php` + the harness adapters (`app/Harness/Adapters/*Adapter.php`) that build the NormalizedSession.
3. **Store project identity:** persist each Project's repo identity (new column, e.g. `projects.repo_identity`, computed from its `root_path` git remote) so resolution is a cheap indexed lookup, not a per-session shell-out.
4. **Resolver fallback** (`SessionRegistryResolver`): keep worktree-path + project-path matching as the **primary/fast path** (correct when the checkout hasn't moved; preserves worktree precedence, no regression). When path match fails, fall back to **repo-identity match** (`session.repo_identity == project.repo_identity`) so relocated sessions attribute to the current project.
5. **Backfill / re-attribution:** a command (or extend `ReconcileIngestState`) that re-resolves already-ingested sessions and re-points those whose attribution changes under repo-identity. Best-effort: use stored session repo_identity if present; else derive from the on-disk repo when the old path still exists; else skip.

## Key constraint / honest edge (validated 2026-07-04)
The motivating session is the **hard case**: `~/Herd/lsnet` is **already gone from disk** (verified — relocated to `~/Documents/code/tangols`, remote `git@github.com:mferrara/tangols.git` = `mferrara/tangols`). Its repo identity cannot be derived by shelling the vanished path now. Recoverable only if (a) flower captured identity at original ingest, or (b) the CC `.jsonl` carries git-remote metadata (worker: check whether `ClaudeAdapter` exposes a remote — the resolver currently takes only cwd). Otherwise this 1 historical session may remain unattributed — **acceptable**, since the operator confirmed tangols's "none" is mostly genuine sparsity. The feature's real value is all **future + still-on-disk** relocations.

## Edge cases
- Distinct repos, same basename (e.g. `./lsnet` ≠ `./tangols` per code catalog) → key on **remote URL/host**, never directory basename.
- Repos with no remote → root-commit-hash identity, else stay path-based.
- Worktrees / multiple checkouts of one repo → `worktree.path` match keeps precedence; repo-identity is last-resort only.
- cwd in a subdir of the repo → resolve to git toplevel before deriving identity.
- Non-git cwds → unchanged (path-based).

## Acceptance
- New sessions under any cwd of a project's repo attribute to that project even after the checkout moves.
- Historical sessions whose **old path still exists** re-attribute on backfill.
- Path-based resolution unchanged when valid (no regression; worktree precedence intact); no cross-repo mis-attribution (distinct remotes stay distinct).
- tangols: documented best-effort — attributes if identity recoverable from stored/session data; else acknowledged as genuinely sparse.

## Prior art / touch points
`SessionRegistryResolver` (resolver), `GitInspector::remoteUrl` / `::ownerRepo` (identity, reference impl), `IngestSession` + harness adapters (capture), `CommitIngestService` / `IngestCommits` (repo-keyed reference), `ReconcileIngestState` (backfill host). Related: todo #340 (ingest core), scratchpad 1007 (umbrella multi-project SPEC), brief #153 (cwd mis-attribution, complete). Cross-link (NOT a dependency): **#196** cross-project support (PARKED) — this is the distinct "attribution across a path move" slice.

_Status → planned. Refine stops here; orchestrator owns dispatch._
[12] status_change flower-refine: (no body)

Recommended linked context:
{
    "todos": [],
    "scratchpads": []
}

Execution notes:
- Treat the brief as the source of truth.
- Keep work scoped to this dispatch request.
- Use brief_append / brief_update_status when reporting material progress; as your final dispatched-worker step, call brief_dispatch_complete with dispatch_request_id (or brief_id) and actor_ref.
- Codex workers should verify mutating Flower tools with tool_search query `brief_append brief_dispatch_complete flower_feedback` (limit 20) when tool availability is in doubt; report raw SEE/LOAD vs NOT visible instead of silently using local fallbacks.
- Add a git commit trailer `Brief: #210` to every commit for this brief so flower can exact-link commits back to the brief.