Batch: agent work from agent-develop #92

Merged

EduarD3V merged 33 commits from agent-develop into develop 2026-10-09 09:19:41 +02:00

Conversation 1 Commits 33 Files changed 50 +1918 -485

gotoi-agent commented 2026-10-08 09:24:58 +02:00

Collaborator

Copy link

One batch of agent work, integrated on agent-develop and conflict-free against develop.

Commits:

78b750e Merge pull request 'Sync develop into agent-develop' (#84) from develop into agent-develop

Merged issue PRs:
Merge pull request 'Sync develop into agent-develop' (#84) from develop into agent-develop

CI runs on this PR: the five required contexts on develop must be green before it is merged.
Every contained PR was already green on its own branch.

Refs (#26 etc. - no closing keywords).

One batch of agent work, integrated on `agent-develop` and conflict-free against `develop`. Commits: ``` 78b750e Merge pull request 'Sync develop into agent-develop' (#84) from develop into agent-develop ``` Merged issue PRs: Merge pull request 'Sync develop into agent-develop' (#84) from develop into agent-develop CI runs on this PR: the five required contexts on `develop` must be green before it is merged. Every contained PR was already green on its own branch. Refs (#26 etc. - no closing keywords).

gotoi-agent added 1 commit 2026-10-08 09:25:02 +02:00

Merge pull request 'Sync develop into agent-develop' (#84) from develop into agent-develop 78b750ea1e

gotoi-agent added 13 commits 2026-10-08 14:35:50 +02:00

middleware: typed control-chat object tools replace read_database_table ... 2afb694284

The single untyped read_database_table table browser is replaced by
per-object tools that project explicit fields, page with an opaque
cursor, and enumerate valid fields/statuses on a bad request:
plan_read, rule_read, incident_read, executions_for_plan, execution_read.
read_machine_context is renamed machine_read.

Writes gain plan_update, plan_delete, plan_test, rule_update and
rule_delete: all blocked in plan mode, operator-request gated, and
authorized per row via inScope. plan_test reports the dry run
(compatibility + kill switch + concrete commit) without calling
decide(), so it never enqueues and never leaks a lease.

middleware/src/ai-table-reader.ts is deleted; hub publishes remove
events for plan/rule deletes through the new optional hubRemove dep.

Refs #85

test: cover typed control-chat object tools and gate write tools by operator request ... 1542a81def

(Refs #85)

docs: UC-MW-136 covers typed control-chat object tools ... 3d3e31ab83

Rewrites the use case around per-object reads and writes, updates the
tool lists in UC-MW-151/UC-MW-169, the catalog index, and the two
plan documents that named read_database_table / read_machine_context.

Refs #85

test: pin plan_read summary totals and cursor to the store size ... 8907bb3544

(Refs #85)

test: use value-pinning matchers in the control object tool tests ... 934bc70589

(Refs #85)

review(middleware): restore ControlScope.denial member dropped by the typed-tools rewrite ... aa1bd25d09

The ControlScope interface hunk in 2afb694 replaced the denial member with
canWrite, but four call sites still use scope.denial (runOne, run_on_machines,
dispatch) and resolveControlScope still returns it — tsc reported TS2339/TS2353
on control-agent.ts. Refs #85

review(docs): align UC-MW-136 and the permissions-audit plan with the typed tool surface ... 6369ff0070

UC-MW-136 claimed every listed write runs blockInPlanMode +
requireOperatorRequest; plan_test runs neither (it is filtered from the
plan-mode tool list and has no operator-request gate), so the story and main
flow now separate it from the mutating writes. The permissions-audit plan
still told the reader to edit ai-table-reader.ts / readDatabaseTable, both
deleted by this issue — Task 3a and its test step now describe ControlRowVisibility
and the plan_read visibility check. Refs #85

middleware: compact object-tool descriptions to fit the 2048-token context window ... 086f5aa50c

Refs #85

Revert "middleware: compact object-tool descriptions to fit the 2048-token context window" ... 355a5f3b23

This reverts commit 086f5aa50c.

middleware: budget-gate the control write tools out of a tiny-window harness ... aabfcbdb98

Refs #85

review(middleware): prompt names the write tools the budget gate dropped ... 90625c2f3a

The budget gate advertised 14 of 21 tools on a tiny window while the
system prompt still told the model it operated through all of them, so
the first write attempt burned a turn on a harness tool error — in the
very window the gate exists to protect. The prompt now states which
tier was dropped; reads and the gate itself are unchanged.

Refs #85

review(test): pin the write-tier budget gate contract ... 24623ed0cb

fitControlTools had no direct assertions — only the end-to-end UC-MW-115
tiny-window chat. Two tests now pin the contract: a 2048-token window
drops exactly the seven write tools and keeps every read tool, and a
fitting or unknown window returns the tool list untouched.

Refs #85

Merge pull request 'Issue #85: typed object tools replace read_database_table (budget-gated writes)' (#94) from issue/85-typed-object-tools into agent-develop 0a3e3553ee

gotoi-agent added 11 commits 2026-10-08 19:46:47 +02:00

test: repro false-green scoring + abort status (red) ... 8f229afa4c

Refs #86

middleware: step exit codes override claimed outcome; add aborted ... a43c77a2ca

Refs #86

middleware: derive status from step exit codes, add aborted, verify exit-code assert (UC-MW-170) ... 105511641d

- applyResult scores from the last step: a failing or aborted run can never be remediated
- ExecutionStatus gains aborted; result enum, terminal lists, overview, and control poll include it
- PlanVerify.expect accepts { command, exit_code }; object form without an integer exit_code is rejected at create/patch
- recheck only compares the string form against PromQL output

Refs #86

agent: workflow abort reports status aborted (UC-AG-68) ... db976282e2

An on_failure/goto/next abort hop sets an abort flag, and the run reports
status aborted instead of remediated, so middleware never scores a stopped
plan as fixed. Refs #86

dashboard: aborted execution status (UC-MW-170) ... ef536fb26f

StatusPill, ExecutionPeek retry, app.css destructive badge, and the types
mirror learn aborted. Refs #86

docs: UC-MW-170 outcome/step truth, UC-AG-68 workflow abort (Refs #86) ... 20a82562a0

New UCs; UC-MW-16 notes rewritten to the derived-status rule; UC-MW-122,
UC-MW-107, UC-AG-12, UC-AG-64 related/versions updated. Refs #86

dashboard: aborted retry, facet, and count on the executions surfaces (UC-MW-170) ... db9ab45221

Refs #86

review(middleware): score any failing step, not only the last (Refs #86) 9aee73b5c8

review(docs): UC-MW-170 any-step truth rule and verify note (Refs #86) 64b673d8ba

merge: restore issue #86 work onto origin/agent-develop (Refs #86) ... 9aaf4913a3

# Conflicts:
#	middleware/src/control-agent.ts

Merge pull request 'Issue #86: Plans report remediated after aborting (false green)' (#95) from issue/86-abort-false-green into agent-develop d6bb5c3909

gotoi-agent added 5 commits 2026-10-08 21:00:42 +02:00

test: repro plan/rule update+delete gaps (red) Refs #87 bf626348a1

feat(plans/rules): refuse plan delete while a match rule references it Refs #87 bde005e0e3

docs(uc): UC-MW-171 refuses plan delete while a rule references it Refs #87 9c0519e357

test(uc-mw-171): plan_delete control tool refuses a referenced plan Refs #87 c3b1fb8b18

Merge pull request 'Issue #87: Refuse plan_delete when a match rule references it' (#96) from issue/87-plan-rule-update into agent-develop b638a135c5

gotoi-agent added 3 commits 2026-10-08 23:25:31 +02:00

test: repro plan/rule guard reject imperative phrasing (red) ... 4d6bd7d59c

Refs #88

feat(control-agent): plan/rule guard accepts imperative phrasing; structured refusal; get_agent_guards ... db0da2ca1d

Guard widened to a request verb (incl. fix, implement, set up, clean up, handle, address,
remediate) plus the plan/rule noun or an anaphor for it, so "fix it and improve it" passes.
requireOperatorRequest now throws a JSON refusal naming the satisfying condition and examples;
new read tool get_agent_guards exposes the state before a call.

UC-MW-172 (plan guard), UC-MW-173 (rule guard); UC-MW-137 updated.

Refs #88

Merge pull request 'Issue #88: plan/rule guards surface explicit-ask' (#97) from issue/88-create-guards-explicit into agent-develop 038c94bbab

EduarD3V merged commit 28058f46a9 into develop 2026-10-09 09:19:41 +02:00

EduarD3V referenced this pull request from a commit 2026-10-09 09:19:45 +02:00

Merge pull request 'Batch: agent work from agent-develop' (#92) from agent-develop into develop

EduarD3V approved these changes 2026-10-09 09:21:15 +02:00

Sign in to join this conversation.

Reviewers

No reviewers

EduarD3V

Labels

Clear labels   

ai-assist

Eligible for the autonomous issue pipeline — the runner only works issues carrying this label

  

needs_attention

Waiting on an operator decision. The runner skips this issue while the label is present; the operator answers with a DECISION: comment and the runner then applies it and clears this label.

  

verified

Operator verified the fix in production — runner closes it and does not re-work it

No labels

ai-assist

needs_attention

verified

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

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

EduarD3V/gotoi!92

No description provided.