feat: add pipelines held in this browser to a project

[camielvs]

Description

A project lives on the backend. A pipeline lives in one browser's IndexedDB. Until now the Pipeline item in a project's + Add menu was disabled, because the only pipeline the resources API accepts is one the backend already holds — and on this branch there is no way to put one there. So the one thing a user can actually make was the one thing they could not add.

This branch closes that gap as things actually are: a project row points at a pipeline in the browser that made it. Nothing is copied.

The alternative was a snapshot of the YAML, which would be shareable but would drift from the pipeline it was taken from the moment anyone edited it. Drift is the worse failure, so a pointer it is — and the page says plainly what that costs rather than pretending.

The row

The entity enum is closed (pipeline | agent_session | document) and a pipeline must name a backend pipeline by UUID. A made-up UUID is accepted by the API, but it would sit in a shared table looking exactly like a real backend pipeline id to every other client and to agents, which would then fail to fetch it. So the row is filed as the only thing it honestly is — one carrying no content of its own — and extra_data records what it names:

entity:     "document"
name:       <pipeline name when it was added>
payload:    {}
extra_data: { kind: "pipeline", storage: "browser", localName, localId? }

The pointer has to live in extra_data, not payload: the list endpoint omits payloads entirely, so a list that needed to know what each row held would cost one request per row.

Resolution: the id first, then the name

The id is tried before the name, which is the opposite of what seems natural. A pipeline's content is keyed by name, and the name finds every pipeline while the id only exists for those the newer editor has touched — so name-first looks right. It is wrong on one ordinary sequence:

1. add "training" to a project
2. rename it to "training-v1"
3. call the next pipeline "training"

By name, the project now resolves to a different pipeline, and looks perfectly successful doing it. Registry ids are never reused, so an id that still resolves is the pipeline that was added. The name stays as the fallback, and a pointer that resolves by neither reports itself as unresolved rather than guessing.

The id is read, never created. Creating one would enter the pipeline into the registry the newer editor builds its folder listing from, so adding a pipeline to a project would quietly change what another page shows.

What this costs, stated plainly

• A pointer is not shareable. Another member — or the same person in another browser — sees a row that cannot be resolved, and the panel says so and says who added it, rather than rendering as broken.
• It can vanish. Delete the pipeline and the row dangles, where a copy would have survived. That is the same coin as never drifting.
• A rename through the older editor breaks it. That path renames content without updating the registry, so both lookups miss. The row then reports itself unresolved rather than resolving wrongly.
• A name can be recycled with no id. A pipeline that never got a registry row, renamed away, with a new pipeline taking its old name, resolves to the new one. There is no content-stable identity available to prevent this — the digest is recomputed on every write.
• These rows count as documents everywhere counts are shown, because the API knows them only by entity and the /projects cards read resource_counts straight off it. The project page counts the same way rather than contradicting the card, so a browser-held pipeline sits under Documents with a Local pipeline badge. Honest, but it does mean the group heading is not the whole truth about its rows.

Worth raising separately: a pipeline_ref member in the entity enum would remove the document fiction entirely, fix ?entity= filtering and the card counts, and costs an enum member plus a PAYLOAD_ENTITIES entry.

Also here

• A document says what kind it is. Any row whose extra_data.kind names something its group heading does not gets that kind as a badge beside its name — Local pipeline here, but a data_sheet document would read Data sheet without the table learning the word. A row with no kind, or one that only repeats its entity, stays bare.

• Removing one no longer offers to delete it. removingDestroys treated "carries no entity_id" as "holds the only copy", which would have given these rows a bin and "deleting it here deletes it for good." It now asks the weaker question — does the row claim to name a pipeline — so even a row whose pointer has become unusable is still removed, not deleted.

• The editor link for a backend pipeline is gone. The editor addresses a pipeline by name out of browser storage, so the link resolved a backend pipeline's name against the wrong store — opening a local namesake where one existed, an empty canvas where none did. Routing to a backend pipeline does not exist yet, so the preview says so instead.

• Judging a spec no longer throws on one too malformed to have an implementation. A pipeline read from browser storage is whatever YAML is there, and declining to judge it beats breaking the panel it is shown in.

• The picker mounts only while open, so opening a project no longer reads browser storage for a list nobody asked for.

Verification

npm run validate and the full suite (2761 tests) green. Unit tests cover the resolution order — including the recycled-name case above as an explicit regression test — the row encoding, the removal wording, the preview's two states, and the picker.

Driven in a real browser against a live backend, in both themes: a pipeline imported through the app's own import path, added through the picker, previewed as its stored YAML with a validity verdict and a working editor link; offered but disabled on reopening the picker; removed with an X while a document keeps its bin; the page's group counts read back and checked against the card's summary for the same project; and a pointer naming a pipeline this browser does not hold rendering as "Not in this browser" with no editor link and no {} leaking from the empty payload.

🤖 Generated with Claude Code

[github-actions]

🎩 Preview

A preview build has been created at: 09-18-feat_local_pipelines_in_projects/826eab1

[camielvs]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• fix(runs): say why the backend refused a run instead of that it did #2794 : 2 dependent PRs (#2796 , #2798 )
• feat(tangent): let the agent name the project, session and pipelines #2792
• feat(projects): fix cancelling, track shared projects, flag unreachable pipelines #2790
• feat(tangent): keep an agent's edits in frame on the canvas #2789
• feat(tangent): bring Tangent onto the dashboard #2787
• feat(tangent): show a project's runs on the tangent page #2786
• fix(projects): show an agent's run on the project it belongs to #2785
• fix(agent): submit a spec that came from the editor's store #2784
• chore: increase max turns in remote agent #2779
• fix(tangent): stop the tab strips scrolling up and down #2777
• feat(projects): make the agents' instructions a project document #2776
• feat(tangent): list the pipelines the backend holds #2775
• feat(projects): show a project's documents on both pages #2774
• fix(tangent): project window parity — share, new-session links, agent-created pipelines, and a folded-away project window #2773
• refactor(projects): describe a project's resources one way #2765
• feat(projects): delete and share a project from either page #2764
• feat: open a new project ready to work in #2763
• feat: let the project page and Tangent reach each other #2762
• fix: keep a long project name inside the tile it belongs to #2761
• feat: say so when Tangent cannot be reached #2760
• feat: open a project where the work happens #2759
• feat: run a pipeline inside a project #2754
• feat: add pipelines held in this browser to a project #2745 👈 (View in Graphite)
• feat: project page #2738
• feat: create a project from a tile in the grid #2735 : 2 other dependent PRs (#2736 , #2737 )
• feat: projects dashboard #2734
• chore: tangent - annotate runs submission to projectid #2758
• fix: tangent - proper debug sequence system prompt #2757
• fix: tangent - quick run button from mini content window #2744
• fix: tangent - remote agent send delta immediately #2743
• fix: tangent - mint proper authorization with tangent #2742
• feat: tangent - debug in tangent sequence #2741
• chore: add autolayout tool #2740
• feat: tangent - resource management via document entity #2733
• feat: tangent - interactive chips in tangent chat #2732
• chore: save id stack to replay ids on reloads #2731
• refactor: tangent - migrate tangetproectcontext onto mobx #2730
• feat: tangent - keep track of open tabs while switching sessions inside project #2729
• feat: tangent - remote agent for runview and editor #2728
• refactor: tangent - structured workarea target #2727
• feat: tangent - embedded pipeline run view #2726
• feat: tangent - embedded pipeline editor #2725
• feat: tangent - dynamic workarea remote tools #2724
• feat: tangent - resources management in project view #2723
• feat: tangent project view page #2722
• feat: workspaces and projects service #2720
• chore: api client for workspaces and projects #2719
• master

This stack of pull requests is managed by Graphite. Learn more about stacking.