Reconcile retained model-opaque Community harness refactor #73

Closed
opened 2026-08-06 03:58:01 +00:00 by coilyco-ops · 5 comments
Member

Preserved branch evidence

  • Branch: feat/sirens-community-harness
  • The branch retains one unique 16-file refactor promoting Echo to a model-opaque Community harness.
  • The branch has no pull request.
  • Current main contains the earlier Community harness work but not this retained promotion as an equivalent patch.

Account for the retained refactor against current main before deleting the branch.

## Preserved branch evidence * Branch: `feat/sirens-community-harness` * The branch retains one unique 16-file refactor promoting Echo to a model-opaque Community harness. * The branch has no pull request. * Current `main` contains the earlier Community harness work but not this retained promotion as an equivalent patch. Account for the retained refactor against current `main` before deleting the branch.
Author
Member

Disposition: recommend closing as superseded

feat/sirens-community-harness now has a pull request, #77, and it does not merge. mergeable: false, 16 files, +192 / -84.

The reason is supersession, not a routine rebase. The merge base is 9a7a3fb dated 2026-07-28. main has advanced 28 commits in the ten days since, and several of them deliver this branch's purpose by a different route:

  • 5e9df6b and #68, 2026-08-04 and 2026-08-05, decoupled Echo from role-intent routes.
  • a033ea6, 2026-08-05, added deployment-selectable response profiles.
  • 5704293, 2026-08-05, named the DeepSeek profile Sirens Deep and added agent/sirens-deep.json.
  • f869390, 2026-08-05, preserved DeepSeek reasoning across tool turns.
  • bcc5998 and 6362e32, 2026-08-06, defined the CoilyCo general-purpose harness and moved harness definitions to YAML.

main is model-opaque today. This branch is the July 28 approach to a problem main solved differently and more completely.

This issue's original framing holds up: the branch content is genuinely absent from main. What changed is the conclusion drawn from that. Absent diff is not unmet purpose.

Before closing

Diff the branch against current main and confirm nothing in those 16 files is uniquely valuable. The strong prior is that everything is superseded, but that is a prior, not a verification, and this issue exists precisely to account for the branch rather than assume.

If the diff is clean, close #77 as superseded, delete the branch, and close this issue with that disposition recorded. If something unique survives, cherry-pick only that.

Do not attempt to resolve the conflicts and merge. Reconciling a ten-day-old parallel approach into a repository that already ships the capability adds risk with no gain, twelve days before coilysiren/inbox#324.

## Disposition: recommend closing as superseded `feat/sirens-community-harness` now has a pull request, #77, and it does not merge. `mergeable: false`, 16 files, +192 / -84. The reason is supersession, not a routine rebase. The merge base is `9a7a3fb` dated 2026-07-28. `main` has advanced 28 commits in the ten days since, and several of them deliver this branch's purpose by a different route: * `5e9df6b` and #68, 2026-08-04 and 2026-08-05, decoupled Echo from role-intent routes. * `a033ea6`, 2026-08-05, added deployment-selectable response profiles. * `5704293`, 2026-08-05, named the DeepSeek profile Sirens Deep and added `agent/sirens-deep.json`. * `f869390`, 2026-08-05, preserved DeepSeek reasoning across tool turns. * `bcc5998` and `6362e32`, 2026-08-06, defined the CoilyCo general-purpose harness and moved harness definitions to YAML. `main` is model-opaque today. This branch is the July 28 approach to a problem `main` solved differently and more completely. This issue's original framing holds up: the branch content is genuinely absent from `main`. What changed is the conclusion drawn from that. Absent diff is not unmet purpose. ## Before closing Diff the branch against current `main` and confirm nothing in those 16 files is uniquely valuable. The strong prior is that everything is superseded, but that is a prior, not a verification, and this issue exists precisely to account for the branch rather than assume. If the diff is clean, close #77 as superseded, delete the branch, and close this issue with that disposition recorded. If something unique survives, cherry-pick only that. Do not attempt to resolve the conflicts and merge. Reconciling a ten-day-old parallel approach into a repository that already ships the capability adds risk with no gain, twelve days before https://forgejo.coilysiren.me/coilysiren/inbox/issues/324.
Author
Member

Diff read: do not merge, and the reason is stronger than supersession

Diffed origin/feat/sirens-community-harness against main at 6362e32. One commit on the branch, 714d718. 28 commits on main since the 9a7a3fb merge base.

The branch adds role-intent routing. main deliberately removed it.

The entire commit replaces the literal AGENT_PROXY_MODEL tag with a logical role/intent route resolved by Agent Proxy:

  • Config.AgentProxyModel becomes Config.AgentProxyRoute plus Config.Intent.
  • Definition gains Harness, Intents, and DefaultIntent.
  • A new Definition.LogicalRoute(intent) returns role + "/" + intent, producing keys such as community/knowledge-retrieval.
  • ProxyClient swaps Model/Role/Seat for Route/Intent/Harness, and sends the route in the model field.
  • AGENT_PROXY_MODEL is dropped from required configuration.
  • Schema bumps to sirens-discord-ops.agent.v3, with a hard check that harness == "sirens-discord-ops" and role == "community".

That is exactly what #67 and #68 took out on 2026-08-04 and 2026-08-05 under the title "Remove role-intent routing from Sirens Echo," and it matches the direction recorded in coilysiren/inbox#323, which holds that multi-harness role-intent routing is unproven.

It would also revert the profile architecture. Confirmed on main: no LogicalRoute, no AgentProxyRoute, no DefaultIntent, no route-intent concept anywhere in internal/, cmd/, or agent/. main selects the model through AGENT_PROXY_MODEL and selects behavior through YAML profiles at agent/sirens-echo.yaml and agent/sirens-deep.yaml, schema coilyco-harness.agent.v1. The branch expects JSON definitions at schema sirens-discord-ops.agent.v3 and hard-requires role == "community", which the CoilyCo general-purpose profile is not.

Those two designs cannot both be true in one tree. The conflicts are the architecture disagreeing with itself, not drift.

Close #77 without merging, delete feat/sirens-community-harness, and close this issue recording that the branch was superseded by the profile and YAML-harness work and contradicts the accepted removal of role-intent routing.

Nothing in the 16 files is worth cherry-picking on its own. The Go changes are all in service of the routing model being rejected, and the doc changes describe that model.

If logical routing is ever revisited, it should start from main's profile architecture rather than from this branch.

## Diff read: do not merge, and the reason is stronger than supersession Diffed `origin/feat/sirens-community-harness` against `main` at `6362e32`. One commit on the branch, `714d718`. 28 commits on `main` since the `9a7a3fb` merge base. **The branch adds role-intent routing. `main` deliberately removed it.** The entire commit replaces the literal `AGENT_PROXY_MODEL` tag with a logical `role/intent` route resolved by Agent Proxy: * `Config.AgentProxyModel` becomes `Config.AgentProxyRoute` plus `Config.Intent`. * `Definition` gains `Harness`, `Intents`, and `DefaultIntent`. * A new `Definition.LogicalRoute(intent)` returns `role + "/" + intent`, producing keys such as `community/knowledge-retrieval`. * `ProxyClient` swaps `Model`/`Role`/`Seat` for `Route`/`Intent`/`Harness`, and sends the route in the `model` field. * `AGENT_PROXY_MODEL` is dropped from required configuration. * Schema bumps to `sirens-discord-ops.agent.v3`, with a hard check that `harness == "sirens-discord-ops"` and `role == "community"`. That is exactly what #67 and #68 took out on 2026-08-04 and 2026-08-05 under the title "Remove role-intent routing from Sirens Echo," and it matches the direction recorded in https://forgejo.coilysiren.me/coilysiren/inbox/issues/323, which holds that multi-harness role-intent routing is unproven. **It would also revert the profile architecture.** Confirmed on `main`: no `LogicalRoute`, no `AgentProxyRoute`, no `DefaultIntent`, no route-intent concept anywhere in `internal/`, `cmd/`, or `agent/`. `main` selects the model through `AGENT_PROXY_MODEL` and selects behavior through YAML profiles at `agent/sirens-echo.yaml` and `agent/sirens-deep.yaml`, schema `coilyco-harness.agent.v1`. The branch expects JSON definitions at schema `sirens-discord-ops.agent.v3` and hard-requires `role == "community"`, which the CoilyCo general-purpose profile is not. Those two designs cannot both be true in one tree. The conflicts are the architecture disagreeing with itself, not drift. ## Recommended disposition Close #77 without merging, delete `feat/sirens-community-harness`, and close this issue recording that the branch was superseded by the profile and YAML-harness work and contradicts the accepted removal of role-intent routing. Nothing in the 16 files is worth cherry-picking on its own. The Go changes are all in service of the routing model being rejected, and the doc changes describe that model. If logical routing is ever revisited, it should start from `main`'s profile architecture rather than from this branch.
Author
Member

Correction to advice given elsewhere in the 2026-08-10 session.

While auditing stray branches I reported that feat/sirens-community-harness, codex/sirens-echo-deepseek-profile-20260805, feat/thread-scoped-replies, and issue-67 were orphaned pre-squash lineage and safe to prune. The evidence was that git merge-base origin/main <branch> returns nothing for each, and that diffing one against main shows roughly 2974 deletions.

That evidence is accurate and the conclusion drawn from it was too strong. No shared history does not mean no unique content, which is exactly what this issue records for feat/sirens-community-harness: one retained 16-file refactor that main does not contain as an equivalent patch.

So the branch-level advice stands only for branches already reconciled. feat/sirens-community-harness should not be deleted until this issue is resolved, and the other three deserve the same read before anyone prunes them rather than a merge-base check alone.

Confirmed safe to delete, since they are merged and carry nothing unique: docs/drop-private-image-claim, fix/policy-check-image-context, sirens-deep-admission-hardening, and feat/prompt-snapshots.

Correction to advice given elsewhere in the 2026-08-10 session. While auditing stray branches I reported that `feat/sirens-community-harness`, `codex/sirens-echo-deepseek-profile-20260805`, `feat/thread-scoped-replies`, and `issue-67` were orphaned pre-squash lineage and safe to prune. The evidence was that `git merge-base origin/main <branch>` returns nothing for each, and that diffing one against `main` shows roughly 2974 deletions. That evidence is accurate and the conclusion drawn from it was too strong. **No shared history does not mean no unique content**, which is exactly what this issue records for `feat/sirens-community-harness`: one retained 16-file refactor that `main` does not contain as an equivalent patch. So the branch-level advice stands only for branches already reconciled. `feat/sirens-community-harness` should not be deleted until this issue is resolved, and the other three deserve the same read before anyone prunes them rather than a merge-base check alone. Confirmed safe to delete, since they are merged and carry nothing unique: `docs/drop-private-image-claim`, `fix/policy-check-image-context`, `sirens-deep-admission-hardening`, and `feat/prompt-snapshots`.
Author
Member

Approved: close unmerged, delete the branch, close this issue

Direction from Kai, 2026-08-12 session.

The disposition recommended in the two comments above is accepted in full:

  1. Close PR #77 without merging.
  2. Delete feat/sirens-community-harness.
  3. Close this issue recording that the branch was superseded by the profile and YAML-harness work, and that it contradicts the accepted removal of role-intent routing in #67 and #68.

Nothing is cherry-picked. The Go changes are all in service of the routing model that was rejected, and the doc changes describe that model.

The retained refactor is now accounted for, which is what this issue existed to do. Absent diff was not unmet purpose.

If logical routing is ever revisited, it starts from main's profile architecture rather than from this branch.

Note on the third comment

The correction about branch pruning stands and outlives this issue. codex/sirens-echo-deepseek-profile-20260805, feat/thread-scoped-replies, and issue-67 still deserve a content read rather than a merge-base check before anyone prunes them. That is not tracked here — file it separately if it matters.

## Approved: close unmerged, delete the branch, close this issue Direction from Kai, 2026-08-12 session. The disposition recommended in the two comments above is accepted in full: 1. Close PR #77 without merging. 2. Delete `feat/sirens-community-harness`. 3. Close this issue recording that the branch was superseded by the profile and YAML-harness work, and that it contradicts the accepted removal of role-intent routing in #67 and #68. Nothing is cherry-picked. The Go changes are all in service of the routing model that was rejected, and the doc changes describe that model. The retained refactor is now accounted for, which is what this issue existed to do. Absent diff was not unmet purpose. If logical routing is ever revisited, it starts from `main`'s profile architecture rather than from this branch. ## Note on the third comment The correction about branch pruning stands and outlives this issue. `codex/sirens-echo-deepseek-profile-20260805`, `feat/thread-scoped-replies`, and `issue-67` still deserve a content read rather than a merge-base check before anyone prunes them. That is not tracked here — file it separately if it matters.
Author
Member

Accounted for. Nothing to port.

The branch shares no history with main

$ git merge-base origin/main origin/feat/sirens-community-harness
fatal: no merge base

It is an orphan lineage, not a divergence. Its commits carry the earlier coily and coilyco-ai ancestry (lockdown: regenerate from coily release, Sync catalog-block check from coilyco-ai canonical), so it predates the history main grew from. That is why it cannot be diffed as a patch and why "reconcile" has to mean comparing behavior rather than comparing trees.

Tip is 714d718, 2026-07-28, and it is the 16-file refactor this issue names.

What the retained refactor actually does

Reading 714d718 rather than its subject line, the promotion is a routing change:

  • Config.AgentProxyModel becomes Config.AgentProxyRoute
  • the definition schema goes sirens-discord-ops.agent.v2 to v3 and gains harness, role, intents, and default_intent
  • SIRENS_ECHO_INTENT selects an intent, and definition.LogicalRoute(intent) derives the route from it
  • AGENT_PROXY_MODEL stops being required

So model opacity is achieved there by having the definition declare a set of intents and computing a logical route from the selected one.

Why it must not be merged

That is role-intent routing, which #67 and #68 deliberately removed and which #76 records as the reason PR #77 is not a prerequisite for anything. Merging it would reintroduce the mechanism the profile architecture was built to replace.

main already has the outcome, by a different mechanism

The goal of the refactor was that the harness names no model. main satisfies that:

  • no tracked definition carries a model field at all, verified across agent/*.yaml
  • the route is AGENT_PROXY_MODEL, read from the environment at runtime, so the deployment owns it and the repository never states it
  • the definition schema is coilyco-harness.agent.v1, which has no model, intent, or route field to state one with

main is model-opaque because a definition cannot name a model. The branch got there by letting the definition declare intents and derive a route. The stricter of the two is the one already shipped.

Verdict

The retained refactor is superseded, not lost. Its outcome is present on main and its mechanism is one the repository has since rejected on purpose. There is nothing to port.

On deleting the branch

I have not deleted it. Branch deletion is destructive and outside what I take on my own, and the tip is the only place that lineage exists. It is safe to delete on this analysis whenever you want it gone.

Worth noting separately: eighteen other branches are still on the remote, several clearly merged. If a sweep is wanted that is its own issue rather than a side effect of this one.

## Accounted for. Nothing to port. ### The branch shares no history with `main` ``` $ git merge-base origin/main origin/feat/sirens-community-harness fatal: no merge base ``` It is an orphan lineage, not a divergence. Its commits carry the earlier `coily` and `coilyco-ai` ancestry (`lockdown: regenerate from coily release`, `Sync catalog-block check from coilyco-ai canonical`), so it predates the history `main` grew from. That is why it cannot be diffed as a patch and why "reconcile" has to mean comparing behavior rather than comparing trees. Tip is `714d718`, 2026-07-28, and it is the 16-file refactor this issue names. ### What the retained refactor actually does Reading `714d718` rather than its subject line, the promotion is a **routing change**: * `Config.AgentProxyModel` becomes `Config.AgentProxyRoute` * the definition schema goes `sirens-discord-ops.agent.v2` to `v3` and gains `harness`, `role`, `intents`, and `default_intent` * `SIRENS_ECHO_INTENT` selects an intent, and `definition.LogicalRoute(intent)` derives the route from it * `AGENT_PROXY_MODEL` stops being required So model opacity is achieved there by having the definition declare a set of intents and computing a logical route from the selected one. ### Why it must not be merged That is role-intent routing, which #67 and #68 deliberately removed and which #76 records as the reason PR #77 is not a prerequisite for anything. Merging it would reintroduce the mechanism the profile architecture was built to replace. ### `main` already has the outcome, by a different mechanism The goal of the refactor was that the harness names no model. `main` satisfies that: * no tracked definition carries a model field at all, verified across `agent/*.yaml` * the route is `AGENT_PROXY_MODEL`, read from the environment at runtime, so the deployment owns it and the repository never states it * the definition schema is `coilyco-harness.agent.v1`, which has no model, intent, or route field to state one with `main` is model-opaque because a definition **cannot** name a model. The branch got there by letting the definition declare intents and derive a route. The stricter of the two is the one already shipped. ### Verdict The retained refactor is superseded, not lost. Its outcome is present on `main` and its mechanism is one the repository has since rejected on purpose. There is nothing to port. ### On deleting the branch I have not deleted it. Branch deletion is destructive and outside what I take on my own, and the tip is the only place that lineage exists. It is safe to delete on this analysis whenever you want it gone. Worth noting separately: eighteen other branches are still on the remote, several clearly merged. If a sweep is wanted that is its own issue rather than a side effect of this one.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
coilyco-gaming/sirens-echo#73
No description provided.