nsaspy/prolog-rlm
1
0
Fork 0
Code Issues 83 Pull requests 10 Projects Releases Packages Wiki Activity Actions

Bundle AgentProlog + DeepSeek harness in prolog-rlm #132

Closed
lost-rob0t wants to merge 14 commits from agent/127-agentprolog-config into main
pull from: agent/127-agentprolog-config
merge into: nsaspy:main
Conversation 3 Commits 14 Files changed 15 +3016 -8
lost-rob0t commented 2026-08-19 17:24:20 +00:00 (Migrated from github.com)
Copy link

Closes #127. Implements the current in-repo product direction tracked by #184.

Architectural correction — 2026-08-27

This PR is the active implementation base for the user-approved repository boundary:

lost-rob0t/prolog-rlm
├── prolog/                 reusable RLM runtime/library
├── agentProlog/            bundled application + trusted operator config
├── harness/deepseek_tui/   bundled reference harness/frontend
└── flake.nix               library + application + harness outputs

There is no separate downstream AgentProlog product owner in the target architecture, and #184 is no longer an optional independent official-DSH product track. #141 is superseded.

The library/application dependency rule remains strict:

harness -> prolog_agent_ui_v1 -> AgentProlog -> public prolog-rlm APIs

Nothing under reusable prolog/ may import frontend/product code. One repository does not mean one monolithic runtime.

Current implementation

The branch now reuses the already-existing reusable prolog_agent_ui_v1 + prolog_agent_ui_facade boundary rather than inventing a second harness transport.

Added in this recovery:

  • agentProlog/prolog/agentprolog_ui.pl — application adapter over the canonical UI protocol;
  • bin/agentprolog-ui.pl — persistent UTF-8 NDJSON protocol server;
  • run.submit — starts one canonical rlm_async Future around the AgentProlog/RLM execution path;
  • session.poll — observes that Future without frontend-owned execution state;
  • session.cancel — cancels the canonical Future and linked child work;
  • deterministic deferred-Future tests for negotiation, request correlation, exactly-once terminal projection and cancellation;
  • apps.agentprolog-ui and packaged protocol smoke in the flake;
  • a persistent Go prolog_agent_ui_v1 client for the bundled Bubble Tea v2 frontend;
  • Esc cancellation from the TUI to canonical Prolog execution;
  • deepseek-harness --check now performs a real Go -> packaged agentprolog-ui negotiation and clean shutdown;
  • a committed Go dependency lockfile for the harness.

AgentProlog application

bin/agentprolog.pl and agentProlog/prolog/agentprolog_cli.pl remain a thin application facade over canonical rlm_cli behavior.

The application does not duplicate:

  • provider/model execution;
  • planning/recursion/subagents;
  • prompt/context compilation;
  • tools/capabilities/authority/effects;
  • history/memory;
  • verification/repair;
  • tracing/usage semantics.

Provider profiles and trusted operator configuration are application policy only.

Programmable configuration

agentprolog_config remains the single AgentProlog operator/configuration authority:

  • trusted executable XDG config.prolog;
  • JSON through the same resolution path;
  • generation-aware reload;
  • project trust gating;
  • frontend-visible projection;
  • secret-like setting rejection;
  • privileged atomic writes.

Configuration does not grant tool capability/effect authority and must remain compatible with generic scoped-state/project-policy boundaries.

Salvage rule from old PR #125

Do not revive #125 wholesale. Port only useful behavioral contracts with fresh TDD against current APIs.

Salvaged/already represented:

  • protocol/handshake versioning;
  • request/session correlation;
  • cancellation reaching canonical Prolog execution;
  • persistent process boundary;
  • focused harness integration checks.

Still to harden before leaving draft:

  • explicit post-negotiation child-process crash regression;
  • canonical runtime status/usage projection in the interactive harness;
  • exact-head CI/Nix evidence.

Explicitly do not restore:

  • the upstream DSH submodule/full source build;
  • custom DSH headless runner;
  • broad Cordis/profile disable patches;
  • private deepseek-harness.json settings;
  • any second provider/tool/history/planner/authority runtime.

Flake outputs

packages.prolog-rlm
packages.default            # reusable prolog-rlm library/package
packages.agentprolog
packages.deepseek-harness

apps.swipl
apps.agentprolog
apps.agentprolog-ui
apps.deepseek-harness

Verification status

  • architecture/issue state reconciled around #184;
  • canonical existing UI protocol reused rather than duplicated;
  • deterministic request/cancel lifecycle tests added;
  • bundled Go TUI switched off one-shot CLI stdout scraping;
  • real packaged Go <-> Prolog negotiation check added;
  • missing Go checksum lockfile identified from Actions and committed;
  • pin actual Nix vendorHash if the next fixed-output build requests it;
  • exact-head deterministic CI green;
  • exact-head Nix flake check green;
  • clean SWI pack + Tree-sitter lanes green;
  • live-provider lanes reviewed separately where credentials exist;
  • explicit post-negotiation process-crash regression;
  • canonical runtime status/usage surfaced in the interactive harness.

Do not weaken deterministic CI or claim old-head evidence for a moved head.

Related

  • #184 — canonical bundled-product/harness migration epic
  • #183 — canonical permanent RLM context/prompt compiler ownership
  • #219 — retrieval/source-aware RLM work remains reusable runtime functionality
  • #141 — superseded standalone downstream-product architecture
  • #231 — old closure rationale superseded; useful generic status work remains valid
Closes #127. Implements the current in-repo product direction tracked by #184. ## Architectural correction — 2026-08-27 This PR is the active implementation base for the user-approved repository boundary: ```text lost-rob0t/prolog-rlm ├── prolog/ reusable RLM runtime/library ├── agentProlog/ bundled application + trusted operator config ├── harness/deepseek_tui/ bundled reference harness/frontend └── flake.nix library + application + harness outputs ``` There is **no separate downstream AgentProlog product owner** in the target architecture, and #184 is no longer an optional independent official-DSH product track. #141 is superseded. The library/application dependency rule remains strict: ```text harness -> prolog_agent_ui_v1 -> AgentProlog -> public prolog-rlm APIs ``` Nothing under reusable `prolog/` may import frontend/product code. One repository does not mean one monolithic runtime. ## Current implementation The branch now reuses the already-existing reusable `prolog_agent_ui_v1` + `prolog_agent_ui_facade` boundary rather than inventing a second harness transport. Added in this recovery: - `agentProlog/prolog/agentprolog_ui.pl` — application adapter over the canonical UI protocol; - `bin/agentprolog-ui.pl` — persistent UTF-8 NDJSON protocol server; - `run.submit` — starts one canonical `rlm_async` Future around the AgentProlog/RLM execution path; - `session.poll` — observes that Future without frontend-owned execution state; - `session.cancel` — cancels the canonical Future and linked child work; - deterministic deferred-Future tests for negotiation, request correlation, exactly-once terminal projection and cancellation; - `apps.agentprolog-ui` and packaged protocol smoke in the flake; - a persistent Go `prolog_agent_ui_v1` client for the bundled Bubble Tea v2 frontend; - `Esc` cancellation from the TUI to canonical Prolog execution; - `deepseek-harness --check` now performs a real Go -> packaged `agentprolog-ui` negotiation and clean shutdown; - a committed Go dependency lockfile for the harness. ## AgentProlog application `bin/agentprolog.pl` and `agentProlog/prolog/agentprolog_cli.pl` remain a thin application facade over canonical `rlm_cli` behavior. The application does not duplicate: - provider/model execution; - planning/recursion/subagents; - prompt/context compilation; - tools/capabilities/authority/effects; - history/memory; - verification/repair; - tracing/usage semantics. Provider profiles and trusted operator configuration are application policy only. ## Programmable configuration `agentprolog_config` remains the single AgentProlog operator/configuration authority: - trusted executable XDG `config.prolog`; - JSON through the same resolution path; - generation-aware reload; - project trust gating; - frontend-visible projection; - secret-like setting rejection; - privileged atomic writes. Configuration does not grant tool capability/effect authority and must remain compatible with generic scoped-state/project-policy boundaries. ## Salvage rule from old PR #125 Do not revive #125 wholesale. Port only useful behavioral contracts with fresh TDD against current APIs. Salvaged/already represented: - protocol/handshake versioning; - request/session correlation; - cancellation reaching canonical Prolog execution; - persistent process boundary; - focused harness integration checks. Still to harden before leaving draft: - explicit post-negotiation child-process crash regression; - canonical runtime status/usage projection in the interactive harness; - exact-head CI/Nix evidence. Explicitly do **not** restore: - the upstream DSH submodule/full source build; - custom DSH headless runner; - broad Cordis/profile disable patches; - private `deepseek-harness.json` settings; - any second provider/tool/history/planner/authority runtime. ## Flake outputs ```text packages.prolog-rlm packages.default # reusable prolog-rlm library/package packages.agentprolog packages.deepseek-harness apps.swipl apps.agentprolog apps.agentprolog-ui apps.deepseek-harness ``` ## Verification status - [x] architecture/issue state reconciled around #184; - [x] canonical existing UI protocol reused rather than duplicated; - [x] deterministic request/cancel lifecycle tests added; - [x] bundled Go TUI switched off one-shot CLI stdout scraping; - [x] real packaged Go <-> Prolog negotiation check added; - [x] missing Go checksum lockfile identified from Actions and committed; - [ ] pin actual Nix `vendorHash` if the next fixed-output build requests it; - [ ] exact-head deterministic CI green; - [ ] exact-head Nix flake check green; - [ ] clean SWI pack + Tree-sitter lanes green; - [ ] live-provider lanes reviewed separately where credentials exist; - [ ] explicit post-negotiation process-crash regression; - [ ] canonical runtime status/usage surfaced in the interactive harness. Do not weaken deterministic CI or claim old-head evidence for a moved head. ## Related - #184 — canonical bundled-product/harness migration epic - #183 — canonical permanent RLM context/prompt compiler ownership - #219 — retrieval/source-aware RLM work remains reusable runtime functionality - #141 — superseded standalone downstream-product architecture - #231 — old closure rationale superseded; useful generic status work remains valid
lost-rob0t commented 2026-08-25 00:18:35 +00:00 (Migrated from github.com)
Copy link

Trusted configuration dependency from #183

The approved RLM context ADARD is now #183: https://github.com/lost-rob0t/prolog-rlm/issues/183

This config runtime is the natural trusted-host surface for future/default-profile controls such as enabling/disabling the bundled permanent RLM operating skills. Preserve the authority boundary from #183:

  • permanence/default activation is host policy, not skill-authored metadata;
  • global/per-skill disable is a trusted configuration action;
  • changing visibility never grants tool capability/effect authority;
  • config should point the canonical prompt compiler at policy, not create another skill/context selector.

No need to broaden this PR unless the approved implementation slice explicitly includes those settings; this comment records the dependency/ownership boundary.

Workflow gate

Normal ARARD flow is the default. Auto-RAGE is not the default and RAGE may begin only after an explicit design approval. Any new configuration semantics outside an approved design return to ARARD first.

## Trusted configuration dependency from #183 The approved RLM context ADARD is now **#183**: https://github.com/lost-rob0t/prolog-rlm/issues/183 This config runtime is the natural trusted-host surface for future/default-profile controls such as enabling/disabling the bundled permanent RLM operating skills. Preserve the authority boundary from #183: - permanence/default activation is host policy, not skill-authored metadata; - global/per-skill disable is a trusted configuration action; - changing visibility never grants tool capability/effect authority; - config should point the canonical prompt compiler at policy, not create another skill/context selector. No need to broaden this PR unless the approved implementation slice explicitly includes those settings; this comment records the dependency/ownership boundary. ### Workflow gate Normal **ARARD** flow is the default. **Auto-RAGE is not the default and RAGE may begin only after an explicit design approval.** Any new configuration semantics outside an approved design return to ARARD first.
lost-rob0t commented 2026-08-27 04:32:08 +00:00 (Migrated from github.com)
Copy link

Superseded by the 2026-08-27 in-repo product decision

The earlier HOLD rationale in this comment depended on #141's standalone/downstream AgentProlog ownership. That architecture is now explicitly superseded by the user's correction in #184.

Current rule:

  • AgentProlog application/config and the bundled harness live in this repository;
  • reusable prolog/ remains independently consumable and must not import product/frontend code;
  • PR #132 is the active current-main implementation base;
  • #75/#77 still constrain generic scoped-state/project-policy semantics, but they are not a reason to move product composition to another repository;
  • useful old Harness behavior is salvaged into the in-repo product with fresh TDD, not by reviving PR #125.

Continue implementation under #184 and the updated PR body.

## Superseded by the 2026-08-27 in-repo product decision The earlier HOLD rationale in this comment depended on #141's standalone/downstream AgentProlog ownership. That architecture is now explicitly superseded by the user's correction in #184. Current rule: - AgentProlog application/config and the bundled harness live in **this repository**; - reusable `prolog/` remains independently consumable and must not import product/frontend code; - PR #132 is the active current-main implementation base; - #75/#77 still constrain generic scoped-state/project-policy semantics, but they are **not** a reason to move product composition to another repository; - useful old Harness behavior is salvaged into the in-repo product with fresh TDD, not by reviving PR #125. Continue implementation under #184 and the updated PR body.
nsaspy changed title from WIP: Bundle AgentProlog + DeepSeek harness in prolog-rlm to Bundle AgentProlog + DeepSeek harness in prolog-rlm 2026-08-27 20:02:08 +00:00
nsaspy commented 2026-09-02 17:27:18 +00:00
Owner
Copy link

Repository-state correction — 2026-08-31

This PR is closed and unmerged and is no longer an active implementation base.

Canonical authority remains #184, whose body has now been reconciled to current repository state:

  • new AgentProlog/DSH harness work starts from current canonical main;
  • merged prolog_agent_ui_v1 + its facade are the frontend boundary to consume;
  • this PR is a salvage source only for useful tested contracts such as persistent process lifecycle, negotiation/versioning, request/session correlation, exactly-once terminal projection, cancellation propagation, and crash handling;
  • do not revive this branch wholesale and do not name it as the base of a new implementation branch;
  • PR #125 remains salvage-only for the same reason.

Fresh implementation should use TDD against current public/runtime/protocol APIs and keep frontend semantics out of DSH.

## Repository-state correction — 2026-08-31 This PR is **closed and unmerged** and is no longer an active implementation base. Canonical authority remains #184, whose body has now been reconciled to current repository state: - new AgentProlog/DSH harness work starts from current canonical `main`; - merged `prolog_agent_ui_v1` + its facade are the frontend boundary to consume; - this PR is a **salvage source only** for useful tested contracts such as persistent process lifecycle, negotiation/versioning, request/session correlation, exactly-once terminal projection, cancellation propagation, and crash handling; - do not revive this branch wholesale and do not name it as the base of a new implementation branch; - PR #125 remains salvage-only for the same reason. Fresh implementation should use TDD against current public/runtime/protocol APIs and keep frontend semantics out of DSH.
nsaspy closed this pull request 2026-09-04 14:37:22 +00:00

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels   
bug
Something isn't working

  
documentation
Improvements or additions to documentation

  
duplicate
This issue or pull request already exists

  
enhancement
New feature or request

  
good first issue
Good for newcomers

  
help wanted
Extra attention is needed

  
invalid
This doesn't seem right

  
question
Further information is requested

  
wontfix
This will not be worked on

No labels
bug
documentation
duplicate
enhancement
good first issue
help wanted
invalid
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
2 participants
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
nsaspy/prolog-rlm!132
No description provided.