nsaspy/prolog-rlm

Fork 0

[MOVED DOWNSTREAM] DeepSeek Harness integration belongs in lost-rob0t/agentProlog #332

New issue

Closed

opened 2026-09-02 17:25:38 +00:00 by nsaspy · 6 comments

nsaspy commented

2026-09-02 17:25:38 +00:00

Owner

Architecture correction — 2026-08-31

The earlier in-repo AgentProlog/harness ownership written into this issue was wrong.

Canonical product owner: lost-rob0t/agentProlog.

lost-rob0t/prolog-rlm remains the reusable runtime/library and generic protocol provider. The authoritative upstream repository boundary is now prolog-rlm#141.

Correct split

lost-rob0t/agentProlog
  -> DeepSeek Harness / Cordis plugin
  -> AgentFactory
  -> product config / UX
  -> coding workflow + coding tool packs
  -> frontend/editor integrations
  -> public prolog-rlm APIs

lost-rob0t/prolog-rlm
  -> reusable RLM runtime
  -> provider/tool/authority/effect semantics
  -> conversations/context
  -> agents/subagents
  -> Spec/Plan/Verify/repair
  -> MCP/project/source primitives
  -> tracing/usage
  -> generic prolog_agent_ui_v1 protocol/facade

Dependency direction is agentProlog -> prolog-rlm, never the reverse.

Current implementation truth

The old nested prolog-rlm/agentProlog/ harness was removed by merged PR #187.
Generic prolog_agent_ui_v1 protocol/facade code remains in prolog-rlm core.
Closed/unmerged PR #132 is historical salvage only and is not an implementation base.
Standalone lost-rob0t/agentProlog already owns the live DSH work.
lost-rob0t/agentProlog#7 is the active DSH AgentFactory issue.
lost-rob0t/agentProlog PR #8 is the current executable implementation slice.

What remains upstream

Only generic runtime/protocol defects exposed by AgentProlog belong here. Open focused upstream issues when the product needs a reusable core contract changed; do not move product composition back into this repository.

Superseded content

All previous text in this issue claiming that AgentProlog, product configuration, or bundled DSH harnesses should live inside prolog-rlm is superseded.

Product work continues in lost-rob0t/agentProlog. Upstream architecture authority: #141.

## Architecture correction — 2026-08-31 The earlier in-repo AgentProlog/harness ownership written into this issue was wrong. **Canonical product owner: `lost-rob0t/agentProlog`.** `lost-rob0t/prolog-rlm` remains the reusable runtime/library and generic protocol provider. The authoritative upstream repository boundary is now `prolog-rlm#141`. ## Correct split ```text lost-rob0t/agentProlog -> DeepSeek Harness / Cordis plugin -> AgentFactory -> product config / UX -> coding workflow + coding tool packs -> frontend/editor integrations -> public prolog-rlm APIs lost-rob0t/prolog-rlm -> reusable RLM runtime -> provider/tool/authority/effect semantics -> conversations/context -> agents/subagents -> Spec/Plan/Verify/repair -> MCP/project/source primitives -> tracing/usage -> generic prolog_agent_ui_v1 protocol/facade ``` Dependency direction is `agentProlog -> prolog-rlm`, never the reverse. ## Current implementation truth - The old nested `prolog-rlm/agentProlog/` harness was removed by merged PR #187. - Generic `prolog_agent_ui_v1` protocol/facade code remains in `prolog-rlm` core. - Closed/unmerged PR #132 is historical salvage only and is not an implementation base. - Standalone `lost-rob0t/agentProlog` already owns the live DSH work. - `lost-rob0t/agentProlog#7` is the active DSH AgentFactory issue. - `lost-rob0t/agentProlog` PR #8 is the current executable implementation slice. ## What remains upstream Only generic runtime/protocol defects exposed by AgentProlog belong here. Open focused upstream issues when the product needs a reusable core contract changed; do not move product composition back into this repository. ## Superseded content All previous text in this issue claiming that AgentProlog, product configuration, or bundled DSH harnesses should live inside `prolog-rlm` is superseded. Product work continues in `lost-rob0t/agentProlog`. Upstream architecture authority: #141.

nsaspy closed this issue

2026-09-02 17:25:38 +00:00

nsaspy commented

2026-09-02 17:26:07 +00:00

Author

Owner

approved for design

nsaspy commented

2026-09-02 17:26:07 +00:00

Author

Owner

Approval reconciliation

The design was explicitly approved by the user in this issue thread (approved for design). The title has been updated to [ADARD][APPROVED].

Implementation ordering is now:

complete #186 repository cleanup and removal of legacy nested/product harnesses;
then implement this approved plugin-only Harness architecture from the cleaned runtime boundary.

The old PR #125/fork-style Harness path is not the implementation base for this ADARD.

## Approval reconciliation The design was explicitly approved by the user in this issue thread (`approved for design`). The title has been updated to `[ADARD][APPROVED]`. Implementation ordering is now: 1. complete #186 repository cleanup and removal of legacy nested/product harnesses; 2. then implement this approved plugin-only Harness architecture from the cleaned runtime boundary. The old PR #125/fork-style Harness path is not the implementation base for this ADARD.

nsaspy commented

2026-09-02 17:26:07 +00:00

Author

Owner

Current workflow status

The ADARD design is approved by the user (see the existing approved for design comment). The stale pre-approval sentence in the original issue body is superseded by that approval and the current issue title.

Implementation ordering is now:

land #186 / PR #187 and recheck the runtime-only main boundary;
begin the approved #184 plugin-only Harness core from that clean base;
return to a design gate only if implementation evidence requires changing an approved architectural boundary.

Do not revive #124 / PR #125 or a nested agentProlog/ harness as the implementation base.

## Current workflow status The ADARD design is **approved** by the user (see the existing `approved for design` comment). The stale pre-approval sentence in the original issue body is superseded by that approval and the current issue title. Implementation ordering is now: 1. land #186 / PR #187 and recheck the runtime-only `main` boundary; 2. begin the approved #184 plugin-only Harness core from that clean base; 3. return to a design gate only if implementation evidence requires changing an approved architectural boundary. Do not revive #124 / PR #125 or a nested `agentProlog/` harness as the implementation base.

nsaspy commented

2026-09-02 17:26:07 +00:00

Author

Owner

Architecture clarification approved in #231: DeepSeek Harness is the reference IDE/workspace renderer for the bundled prolog-rlm CLI, not the semantic agent runtime and not a separate AgentProlog product owner. Prolog-RLM keeps canonical ownership of context compilation/projection, provider/model calls, tools/capabilities/authority/effects, history/memory, recursion/subagents, project/source KB, retrieval, verification/repair, and usage/status. The DSH adapter should consume prolog_agent_ui_v1; if that protocol lacks required semantics, extend/version the protocol rather than importing runtime internals into DSH. Keep a one-shot prolog-rlm run "task" path alongside prolog-rlm ide .. Canonical status includes model + token input/output + context occupancy percent when known (ctx ? when capacity is unknown). #233 is the first TDD implementation slice.

Architecture clarification approved in #231: DeepSeek Harness is the **reference IDE/workspace renderer for the bundled `prolog-rlm` CLI**, not the semantic agent runtime and not a separate AgentProlog product owner. Prolog-RLM keeps canonical ownership of context compilation/projection, provider/model calls, tools/capabilities/authority/effects, history/memory, recursion/subagents, project/source KB, retrieval, verification/repair, and usage/status. The DSH adapter should consume `prolog_agent_ui_v1`; if that protocol lacks required semantics, extend/version the protocol rather than importing runtime internals into DSH. Keep a one-shot `prolog-rlm run "task"` path alongside `prolog-rlm ide .`. Canonical status includes model + token input/output + context occupancy percent when known (`ctx ?` when capacity is unknown). #233 is the first TDD implementation slice.

nsaspy commented

2026-09-02 17:26:08 +00:00

Author

Owner

Scope clarification — 2026-08-27

PR #132 now adds a small in-repo DeepSeek API harness with a Bubble Tea v2 TUI over the AgentProlog CLI. That does not silently implement or replace this issue's separate official DeepSeek Harness / AgentFactory-plugin integration design.

Current split:

#132: AgentProlog monorepo application layer + direct DeepSeek provider profile + Bubble Tea/Bubbles terminal frontend;
#184: optional future integration with the official deepseek-harness product/plugin architecture, if still wanted.

Both must route semantic execution through Prolog-RLM and neither may create a second settings/provider/authority runtime. Do not revive the old #125 nested Harness implementation as part of #132.

## Scope clarification — 2026-08-27 PR #132 now adds a **small in-repo DeepSeek API harness with a Bubble Tea v2 TUI** over the AgentProlog CLI. That does **not** silently implement or replace this issue's separate official DeepSeek Harness / AgentFactory-plugin integration design. Current split: - #132: AgentProlog monorepo application layer + direct DeepSeek provider profile + Bubble Tea/Bubbles terminal frontend; - #184: optional future integration with the official `deepseek-harness` product/plugin architecture, if still wanted. Both must route semantic execution through Prolog-RLM and neither may create a second settings/provider/authority runtime. Do not revive the old #125 nested Harness implementation as part of #132.

nsaspy commented

2026-09-02 17:26:08 +00:00

Author

Owner

No GPT, it does replace it because we're not gonna use that repo. Everything should be provided from here. It's so simple. I can't believe you can't see it. Of course. To harness work will need to be migrated back to it.

nsaspy referenced this issue

2026-09-02 17:26:35 +00:00

[direct] add an explicit native-call batch cardinality budget (max_native_calls_per_batch) #305

nsaspy referenced this issue

2026-09-02 17:26:35 +00:00

[direct] add an explicit native-call batch cardinality budget (max_native_calls_per_batch) #305