Issue #5: control chat must not create plans or rules unprompted #6

Merged

EduarD3V merged 5 commits from issue/5-chat-no-autoplan into develop 2026-10-05 08:18:08 +02:00

Conversation 1 Commits 5 Files changed 5 +123 -16

EduarD3V commented 2026-10-04 09:20:42 +02:00

Owner

Copy link

Refs #5.

The control-chat assistant could call create_plan / create_match_rule on its own initiative (e.g. to record work it had just done), so a chat session that ended with a plan created one the operator never asked for.

Change: create_plan and create_match_rule now require an explicit operator request in the message that started the current turn (ControlDeps.lastUserMessage, set by runControlAgent). The check is a conservative English guard rail - a request verb (create/make/add/save/build/draft/...) plus the noun plan/playbook/rule - and the tool throws, persisting nothing, otherwise. The tool descriptions state the rule; the model-visible text is net shorter than develop. UC-MW-137 updated before the behaviour change.

Commits:

• 7c2abe9 docs(uc): UC-MW-137 gate AI plan/rule creation on operator request
• 4a47de6 feat(control): create_plan/create_match_rule require an explicit operator request
• 9152969 fix(control): keep plan/rule prompt text within the tiny context window

Gates (touched surface, all green):

• python3 docs/use-cases/_check.py - 350 UCs conform
• python3 scripts/check-ai-headers.py - 534 ok
• python3 scripts/check-type-mirror.py - ok
• bun test tests/uc-mw-137-ai-create-plan-rule.test.ts - 5/5 (2 new refusal tests, additive)
• bun test tests/uc-mw-115-control-chat.test.ts - 12/12
• bun test tests/uc-mw-plan-chat.test.ts - 11/11

Known red, NOT from this diff: the full 'cd middleware && bun test' suite fails ~18 unrelated tests (cross-user isolation, secrets, /metrics, UC-MW-19) with 'PGlite is closed' / ERR_HTTP_HEADERS_SENT. The untouched base (origin/develop 33143a0) fails the same set (541 pass/20 fail vs 545/18 here), so this is pre-existing environment flakiness, not a regression. Opened as draft for that reason only.

Not touched: dashboard (the manual 'Save as plan' button is already an explicit operator action) and /plans/compile (operator-initiated). A human merges into develop.

Refs #5. The control-chat assistant could call create_plan / create_match_rule on its own initiative (e.g. to record work it had just done), so a chat session that ended with a plan created one the operator never asked for. Change: create_plan and create_match_rule now require an explicit operator request in the message that started the current turn (ControlDeps.lastUserMessage, set by runControlAgent). The check is a conservative English guard rail - a request verb (create/make/add/save/build/draft/...) plus the noun plan/playbook/rule - and the tool throws, persisting nothing, otherwise. The tool descriptions state the rule; the model-visible text is net shorter than develop. UC-MW-137 updated before the behaviour change. Commits: - 7c2abe9 docs(uc): UC-MW-137 gate AI plan/rule creation on operator request - 4a47de6 feat(control): create_plan/create_match_rule require an explicit operator request - 9152969 fix(control): keep plan/rule prompt text within the tiny context window Gates (touched surface, all green): - python3 docs/use-cases/_check.py - 350 UCs conform - python3 scripts/check-ai-headers.py - 534 ok - python3 scripts/check-type-mirror.py - ok - bun test tests/uc-mw-137-ai-create-plan-rule.test.ts - 5/5 (2 new refusal tests, additive) - bun test tests/uc-mw-115-control-chat.test.ts - 12/12 - bun test tests/uc-mw-plan-chat.test.ts - 11/11 Known red, NOT from this diff: the full 'cd middleware && bun test' suite fails ~18 unrelated tests (cross-user isolation, secrets, /metrics, UC-MW-19) with 'PGlite is closed' / ERR_HTTP_HEADERS_SENT. The untouched base (origin/develop 33143a0) fails the same set (541 pass/20 fail vs 545/18 here), so this is pre-existing environment flakiness, not a regression. Opened as draft for that reason only. Not touched: dashboard (the manual 'Save as plan' button is already an explicit operator action) and /plans/compile (operator-initiated). A human merges into develop.

EduarD3V added 3 commits 2026-10-04 09:20:42 +02:00

docs(uc): UC-MW-137 gate AI plan/rule creation on operator request 7c2abe9e48

feat(control): create_plan/create_match_rule require an explicit operator request ... 4a47de667c

Gate the control-chat plan/rule tools on the operator's current message so
the assistant stops creating plans unprompted when a chat ends. Tighten the
system prompt and tool descriptions to match. UC-MW-137.

fix(control): keep plan/rule prompt text within the tiny context window ... 9152969e61

The added prompt sentence and long tool descriptions pushed UC-MW-115's
2048-token compaction harness over budget. Restore the original system
prompt line and shorten both tool descriptions; the explicit-request gate
lives in code, so the text only needs to state the rule.

EduarD3V changed title from Issue #5: control chat must not create plans or rules unprompted to WIP: Issue #5: control chat must not create plans or rules unprompted 2026-10-04 09:20:43 +02:00

EduarD3V referenced this pull request 2026-10-04 09:21:15 +02:00

Ai Chat, when ends, it is always creating new plans #5

EduarD3V commented 2026-10-04 09:33:07 +02:00

Author

Owner

Copy link

Correction to the gate note above: the authoritative full-suite run on this branch is 561 pass / 2 fail, not ~18. The earlier larger failure count came from two bun test runs sharing the machine (PGlite contention). The two remaining failures - UC-OP-02 /metrics counters and UC-MW-19 plan survives a second buildApp - also fail on the untouched base origin/develop 33143a0 (541 pass / 20 fail there), so they are pre-existing environment issues, not from this diff.

Correction to the gate note above: the authoritative full-suite run on this branch is 561 pass / 2 fail, not ~18. The earlier larger failure count came from two bun test runs sharing the machine (PGlite contention). The two remaining failures - UC-OP-02 /metrics counters and UC-MW-19 plan survives a second buildApp - also fail on the untouched base origin/develop 33143a0 (541 pass / 20 fail there), so they are pre-existing environment issues, not from this diff.

EduarD3V added 1 commit 2026-10-04 21:51:49 +02:00

update 1d96a04133

EduarD3V added 1 commit 2026-10-04 21:55:39 +02:00

Merge branch 'develop' of https://git.eduard3v.com/EduarD3V/gotoi into issue/5-chat-no-autoplan ... 612ea75179

# Conflicts:
#	.claude/skills/gotoi-issue-flow/scripts/forgejo.py

EduarD3V changed title from WIP: Issue #5: control chat must not create plans or rules unprompted to Issue #5: control chat must not create plans or rules unprompted 2026-10-05 08:17:42 +02:00

EduarD3V merged commit 5dc62a75b2 into develop 2026-10-05 08:18:08 +02:00

EduarD3V referenced this pull request from a commit 2026-10-05 08:18:10 +02:00

Merge pull request 'Issue #5: control chat must not create plans or rules unprompted' (#6) from issue/5-chat-no-autoplan into develop

EduarD3V referenced this pull request 2026-10-05 08:18:29 +02:00

Ai Chat, when ends, it is always creating new plans #5

Sign in to join this conversation.

Reviewers

No reviewers

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

1 participant

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!6

No description provided.