Skip to content

fix(planning): install planning skills by default, call Jira only when a ticket is already there - #254

Merged
mganchas merged 3 commits into
mainfrom
fix/planning-base-strict-jira
Nov 17, 2026
Merged

mganchas merged 3 commits into
mainfrom
fix/planning-base-strict-jira

Conversation

@mganchas

@mganchas mganchas commented Nov 16, 2026 •

Copy link
Copy Markdown
Contributor

Description

Mark planning, planning-spec, and jira-brief stable and base: true, so a default install includes them. planning wins over a generic plan, brainstorm, design, or spec skill only when the engineer did not name another skill for that job. planning calls jira-brief only when the message has a Jira browse URL or the engineer explicitly refers to a Jira ticket. Otherwise it says nothing about Jira. A planning delegation never asks whether a bare token is a ticket.

Opt out with yarn skills --exclude planning/planning,planning/planning-spec,planning/jira-brief --save.

Type of Change

  • New skill
  • Skill improvement/update
  • Bug fix
  • Documentation update
  • Other (please describe):

Skill Details (if adding a new skill)

Provider Name:
Skill Name:
Brief Description:

Checklist

  • I have read the CONTRIBUTING.md guidelines
  • My skill follows the SKILL_TEMPLATE.md format
  • I have tested this skill with an AI agent
  • My skill does not contain any secrets, private keys, or sensitive data
  • I have added appropriate documentation
  • My changes don't break existing skills

Testing

  • yarn audit:skills: 0 errors. Planning description is 1024 characters.
  • yarn lint:changelog: pass.
  • Cross-lab matrix on commit 57a0a7d: 92 runs (23 scenarios x 4 models). 89 pass / 3 fail / 9 divergent-but-pass. The 3 fails are named-skill runs that reported override=yes while correctly keeping the named skill (flag semantics, not behavior). Hard rules held in every scenario. Details below.
Cross-lab matrix (92 runs)

23 scenarios × 4 models (claude-sonnet-5-5-high, gpt-6.1-sol-medium, gemini-3.1-pro-preview-high, kimi-k3-high) = 92 runs. Each subagent read the installed mms-planning SKILL.md in the real mobile/extension checkout and reported only the path decision. Cell = path (+spec/stopped/jira/override when relevant).

Scenario Claude GPT Gemini Kimi
typo-mobile no-trigger no-trigger no-trigger proceed
typo-ext no-trigger no-trigger no-trigger no-trigger
send-mobile full-brief full-brief full-brief full-brief
swap-ext full-brief full-brief full-brief full-brief
vault-open full-brief, stop full-brief, stop full-brief, stop full-brief, stop
keyring-open full-brief, stop full-brief, stop full-brief, stop full-brief, stop
plan-mobile full-brief full-brief full-brief ⚠ proceed
plan-ext full-brief full-brief full-brief full-brief
label-mobile no-trigger no-trigger no-trigger no-trigger
label-ext no-trigger no-trigger no-trigger no-trigger
markets full-brief, spec✓ spec ✓ full-brief, spec✓ full-brief, spec✓
jest proceed no-trigger no-trigger no-trigger
vault-closed short-brief, stop✓ short-brief, stop✓ short-brief, stop✓ short-brief, stop✓
keyring-method full-brief, stop ⚠ short-brief, stop full-brief, stop full-brief, stop
public-method full-brief, stop full-brief, stop full-brief, stop full-brief, stop
cheap-mobile cheap-note full-brief (allowed) cheap-note cheap-note
ticket-maybe cheap-note, jira✓ cheap-note, jira✓ cheap-note, jira✓ cheap-note, jira✓
plan-rive full-brief, jira✗ full-brief, jira✗ full-brief, jira✗ full-brief, jira✗
browse-url cheap-note, jira✓ cheap-note, jira✓ proceed, jira✓ proceed, jira✓
named-skill no-trigger, ⚠override=yes proceed, override=no✓ no-trigger, ⚠override=yes no-trigger, ⚠override=yes
generic-plan full-brief, override=yes short-brief, override=yes full-brief short-brief
household-mobile spec ✓ spec ✓ full-brief, spec✓ full-brief, spec✓
household-ext spec ✓ spec ✓ full-brief, spec✓ full-brief, spec✓

Counts

  • Pass: 89/92 · Fail: 3 · Divergent-but-pass: 9
  • Fails: named-skill on claude/gemini/kimi reported override=yes. However, all three why lines state the named brainstorming skill ran and mms-planning did not win — behavior is correct; the models inverted the flag's semantics (read "override" as "an override happened" rather than "planning overrode the named skill"). Only GPT answered override=no.
  • Divergent, still within the fail rules: plan-mobile kimi (proceeded on an arguably ambiguous ask), keyring-method gpt (short-brief instead of full-brief), markets + household ×7 runs (full-brief with spec=yes instead of path=spec — spec still recommended), cheap-mobile gpt (full-brief, explicitly acceptable).
  • Hard rules held everywhere: mechanical (typo/label/jest) never stopped; vault/keyring/contract changes always stopped for a brief; spec always recommended for markets/household; jira=yes only with an explicit ticket or URL (ticket-maybe/browse-url 8/8).

Verdict

Safe to merge for base install. Core path decisions are correct across all four labs in every scenario: mechanical work never blocks, security/contract surfaces always stop for agreement, large new product areas get a spec recommendation, Jira engages only when referenced, and no lab let mms-planning override an explicitly named skill. One nit worth a follow-up line in SKILL.md: define the override report field ("override=yes only when mms-planning replaces the named skill") — 3/4 models inverted it despite behaving correctly. Not a blocker.

Raw subagent reports (92 runs)
=== typo-ext-gpt ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The fully specified comment-only typo fix changes no behavior, contract, or risk surface and does not trigger planning.
=== typo-mobile-gpt ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The comment-only typo correction is fully specified and mechanical, with no behavior, contract, security, or performance change.
=== typo-mobile-kimi ===
triggered: yes
path: proceed
spec: no
stopped: no
jira: no
override: n/a
why: A comment-only typo fix is mechanical (cannot change behavior, contract, or risk surface), so it proceeds without a brief, spec, or Jira step.
=== swap-ext-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The ask is ambiguous (it names neither the exact target, action, nor expected result) and leaves open product decisions about the swap confirmation UI, so it stops for a brief without a spec since the change is not large, not security-sensitive, and not a contract/API change.
=== swap-ext-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The request leaves UX scope and expected behavior undefined, requiring an agreed brief but not a spec because it introduces neither a new product area nor an undecided cross-system design.
=== typo-ext-kimi ===
triggered: no
path: no-trigger
spec: no
stopped: no
jira: no
override: n/a
why: A comment-only typo fix is mechanical and none of the skill's trigger conditions apply (not ambiguous, not large, no contract/API change, not security- or performance-sensitive, and mms-planning was not requested), so it proceeds without a brief.
=== send-mobile-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The send-flow request leaves product behavior and the expected result undefined, requiring a brief without establishing a large-change spec trigger.
=== send-mobile-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The ask names no target, action, or expected result, so scope is ambiguous and it stops for a brief, but improving the existing send flow is not a new product area or undecided cross-system design, so no spec trigger applies.
=== send-mobile-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Scope is ambiguous because the request does not specify the target, action, or expected result for improving the send flow.
=== typo-mobile-claude ===
triggered: no
path: no-trigger
spec: no
stopped: no
jira: no
override: n/a
why: A comment-only typo fix is mechanical and has no ambiguity, large scope, security surface, performance surface or contract change, so mms-planning does not apply and the edit would go ahead directly.
=== typo-ext-claude ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: A comment-only typo fix is mechanical and fully specified, with no ambiguity, large scope, security or performance risk, or API change, so mms-planning does not apply.
=== send-mobile-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The ask names no target, action or expected result for the send flow, so scope is ambiguous and it stops for a brief, but it is not a new product area, so no spec.
=== swap-ext-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The ask names no concrete target, action or expected result and leaves open product and UX decisions, so it stops for a brief. It isn't a new product area, so it doesn't get a spec.
=== swap-ext-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The request is ambiguous and not fully specified, requiring a brief to clarify scope and expected results, but it is not large enough to trigger a formal planning spec.
=== typo-mobile-gemini ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: Fixing a comment typo is a mechanical change that does not match any of the conditions in the "When to use" section.
=== typo-ext-gemini ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The request is a mechanical typo fix in a comment with no behavior change, which does not meet any trigger conditions for the planning skill.
=== vault-open-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The change touches the vault, a migration and persisted state, and the persisted shape is still undecided. That makes it security-sensitive and ambiguous, so it stops for a brief. It is not large by itself, so no spec.
=== keyring-open-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: no
why: The ask touches keyring/vault unlock and persisted state, so it is security-sensitive, and it leaves the target, action and expected result open, so it needs a full brief rather than a one-line pause. It is not a new product area or a cross-system design, so no spec is recommended.
=== plan-mobile-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The ask doesn't say which settings label to change or what the new text should be, so it is ambiguous but small and not large, which means a brief and no spec.
=== vault-open-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Vault encryption migration is security-sensitive with an undecided persisted contract, requiring a full brief but not a spec because it is not inherently large.
=== vault-open-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Vault seed-phrase storage with a migration is security-sensitive (vault, migrations, persisted state) and the ask is not fully specified — the persisted shape is undecided — so it stops before any edit for the step-4 brief, with no spec because the change is not large.
=== plan-mobile-kimi ===
triggered: yes
path: proceed
spec: no
stopped: no
jira: no
override: n/a
why: Engineer named mms-planning and the fully specified, non-sensitive single-label copy change is mechanical (mms-content-guidelines applies to the copy itself), so it proceeds without a brief, no spec, and no Jira surface.
=== keyring-open-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Unlock and password-change persistence are security-sensitive with unspecified outcomes, requiring a full brief and agreement but not a spec because no large change is established.
=== keyring-open-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The change touches security-sensitive areas like the keyring and persisted state and is not fully specified, requiring a full brief before proceeding.
=== plan-mobile-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Planning was explicitly requested, but the label and replacement copy remain unspecified, requiring a brief without a spec.
=== vault-open-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The change modifies security-sensitive vault persistence and has an undecided scope, requiring a full brief but no spec because vault migrations are not large by themselves.
=== plan-mobile-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The user explicitly requested mms-planning for an ambiguous ask, which stops for a step 4 brief without a spec or Jira check.
=== keyring-open-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: no
why: The ask touches keyring and persisted state, which is security-sensitive, and it is not fully specified, so step 1/4 requires stopping for a brief before any edit, with no spec because keyring and persisted state are not large by themselves.
=== label-mobile-claude ===
triggered: no
path: no-trigger
spec: no
stopped: no
jira: no
override: n/a
why: The request names the target, the action and the result, and it has no ambiguity, large scope, contract or API change, security or performance surface, or plan request, so mms-planning does not apply.
=== label-mobile-gpt ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The request specifies only a button-label change, with no ambiguity, behavior change, contract change, security risk, or performance impact.
=== label-ext-claude ===
triggered: no
path: no-trigger
spec: no
stopped: no
jira: no
override: n/a
why: The request names the target, action and result, and it is a plain copy change with no layout, logic, security, performance, contract or size concern, so none of mms-planning's triggers apply.
=== plan-ext-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The engineer asked for a plan and gave no target, action or expected result for the network controller, so it stops for a brief, and since it isn't a new product area or an undecided cross-system design, no spec is recommended.
=== label-ext-gpt ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The request specifies a copy-only header change with no ambiguity, layout, logic, contract, security, or performance changes.
=== plan-ext-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The engineer explicitly asks for a plan and the ask is not fully specified — it names no target action or expected result for the network controller — so step 4 stops for a scope brief, with no spec trigger since the change is not shown to be large.
=== plan-ext-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: no
why: The explicit plan request triggers planning, but the controller change lacks an action and expected result, requiring a brief without evidence of a large change.
=== label-mobile-gemini ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The request is a fully specified, mechanical copy change that does not meet any of the trigger conditions for the planning skill.
=== label-ext-kimi ===
triggered: no
path: no-trigger
spec: no
stopped: no
jira: no
override: n/a
why: The ask is fully specified (target: home header string, action: change "MetaMask" to "MetaMask Wallet") and is not ambiguous, large, a contract/API change, or security/performance-sensitive, so no mms-planning trigger applies (mms-content-guidelines would still govern the copy edit itself).
=== plan-ext-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The engineer explicitly requested a plan and the scope of the network controller change is not fully specified.
=== label-ext-gemini ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: A simple string change is a mechanical edit that does not meet any of the triggers for planning (not ambiguous, large, security-sensitive, performance-sensitive, or an API change).
=== label-mobile-kimi ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The ask names the target (settings screen Contact support button), the action (change label), and the expected result (Get support), and it is not large, security- or performance-sensitive, a contract/API change, or a request for a plan, so mms-planning is not used before this edit.
=== markets-claude ===
triggered: yes
path: full-brief
spec: yes
stopped: yes
jira: no
override: n/a
why: A new product area that shares accounts, swaps and notifications, with scope not decided, is large, so it stops for a brief that recommends a spec and ends by asking whether to continue with a planning spec, continue from the brief, or discuss more.
=== vault-closed-claude ===
triggered: yes
path: short-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Re-encrypting the vault touches keyring/vault and persisted state, so it is security-sensitive and has to pause for agreement before the first edit. It is fully specified and not large, so it gets a one-line risk brief and no spec.
=== jest-claude ===
triggered: yes
path: proceed
spec: no
stopped: no
jira: no
override: n/a
why: Raising a Jest coverageThreshold is routine CI-only coverage work that doesn't change how jobs or E2E launch, with no security, performance, contract or large-scope risk, so it proceeds without a brief.
=== jest-gpt ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: Raising one existing Jest coverage threshold by one percentage point is a fully specified routine CI-only change with no planning trigger.
=== markets-kimi ===
triggered: yes
path: full-brief
spec: yes
stopped: yes
jira: no
override: no
why: A new prediction-markets product area is large (new product area) and scope is explicitly undecided, so mms-planning stops for a full brief and recommends a spec; no Jira ticket or other named skill applies.
=== vault-closed-gpt ===
triggered: yes
path: short-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Vault re-encryption is security-sensitive and requires agreement before editing, but this fully specified change within existing vault code is not large and does not warrant a spec.
=== vault-closed-kimi ===
triggered: yes
path: short-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Fully specified ask (target, action, expected result named) but vault work is security-sensitive, so it stops for a one-line short brief naming the risk and waits for agreement; not large, so no spec.
=== markets-gpt ===
triggered: yes
path: spec
spec: yes
stopped: yes
jira: no
override: n/a
why: A new product area sharing accounts, swaps, and notifications with undecided scope is a large cross-system change requiring a brief and spec recommendation before implementation.
=== jest-kimi ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The ask is fully specified (raise one file's Jest coverageThreshold by 1pp, no job/E2E launch changes) and none of the mms-planning triggers apply — not ambiguous, not large, not security- or performance-sensitive, and no contract/API change — it is routine CI-only coverage work that proceeds without the skill.
=== markets-gemini ===
triggered: yes
path: full-brief
spec: yes
stopped: yes
jira: no
override: n/a
why: The request introduces a large, ambiguous new product area without a decided scope, requiring a full brief and a spec recommendation before proceeding.
=== jest-gemini ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: n/a
why: The request is a fully specified, routine CI-only change that does not meet any of the triggers for mms-planning.
=== vault-closed-gemini ===
triggered: yes
path: short-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The request is fully specified and not large but modifies the vault which is security-sensitive, requiring a short brief pause before editing.
=== cheap-mobile-claude ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: no
override: n/a
why: The engineer is undecided about keeping the result, which is cheap uncertainty, and send amount validation is not security- or performance-sensitive here. So it proceeds without a brief, and the final summary ends with one line asking the engineer to confirm whether the result is kept.
=== keyring-method-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: A new public method is a contract/API change in a keyring/vault (security-sensitive) area, so it stops for a brief with the agreement question; it is not large, so there is no spec and no Jira reference.
=== public-method-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Adding a public method is a contract/API change, so it stops for a brief with scope and non-goals. It is not large, and it has no security, performance or persisted-state surface, so no spec is recommended.
=== keyring-method-gpt ===
triggered: yes
path: short-brief
spec: no
stopped: yes
jira: no
override: n/a
why: This fully specified keyring/vault API addition requires a short security-risk brief and agreement before editing, but is not large enough for a spec.
=== public-method-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: A new public method on a controller is a contract/API change, which stops for a brief even when fully specified, and it is not large so no spec is recommended.
=== cheap-mobile-kimi ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: no
override: n/a
why: The engineer named mms-planning and stated they have not decided whether they will keep the result, which is cheap uncertainty ("whether the engineer will keep the result"), so the change proceeds without a brief and the final summary ends with one line asking the engineer to confirm whether they will keep it; send amount validation is not security-sensitive (no auth, secrets, keyring, vault, migrations, or persisted state) nor performance-sensitive (not a render or data path), so no stop applies.
=== cheap-mobile-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The desired validation behavior is undecided, so uncertainty is not limited to keeping the result and requires a brief without a spec.
=== public-method-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Adding a public method changes the controller API and requires a brief even when fully specified, but the controller-only scope is not large enough for a spec.
=== public-method-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Adding a public method is a contract or API change that stops for a brief, but does not trigger a spec because the change is not large.
=== cheap-mobile-gemini ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: no
override: n/a
why: The engineer has not decided whether to keep the result for a change that is not security-sensitive or performance-sensitive, which constitutes cheap uncertainty and proceeds without a brief.
=== keyring-method-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Adding a public method to the keyring controller is a contract/API change on a security-sensitive (keyring/vault) surface, so it stops for a brief before any edit, but it is not large so no spec is recommended.
=== keyring-method-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Adding a public method is a contract or API change that requires a full brief, and because the change is not large, no spec is recommended.
=== ticket-maybe-claude ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: yes
override: n/a
why: The ask is fully specified and low-risk, and the only uncertainty is whether EXT-4401 covers it, which is cheap to undo. So it proceeds (after reading the EXT-4401 brief via mms-jira-brief, since the user named the ticket), and the final summary ends with a line asking the engineer to confirm that coverage.
=== plan-rive-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: no
why: The work is performance-sensitive (screen render and animation startup path) and the ask names no fix, so it stops for a step-4 brief ending in the agreement question. It is a single screen, so it is not large and gets no spec.
=== browse-url-kimi ===
triggered: yes
path: proceed
spec: no
stopped: no
jira: yes
override: n/a
why: The ask points to an existing Jira ticket whose acceptance criteria cover the change, so it proceeds without a brief per step 1 (no security/performance surface known), and the browse URL triggers the Jira step via mms-jira-brief.
=== ticket-maybe-gpt ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: yes
override: n/a
why: Only ticket coverage is uncertain, so proceed without approval or a spec, apply the Jira step, and request coverage confirmation in the final summary.
=== plan-rive-kimi ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: Plan request for a performance-sensitive screen (animation/latency on MoneyOnboardingView) that is not fully specified — target and symptom named but the fix approach is an open decision — and the change is not large, so it stops for a step-4 brief with no spec recommendation and no Jira or named-skill override.
=== plan-rive-gpt ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The requested plan concerns performance-sensitive Rive loading with an unspecified remedy, requiring a brief and agreement but not a spec because the change is not large.
=== browse-url-claude ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: yes
override: n/a
why: The request points to ticket MCWP-726 as its scope, so it proceeds without a brief. Jira would not open (the browse extension is not paired and the fetch returned only "Jira"), so the brief was not read. The final summary must say that and ask the engineer to confirm the ticket covers the change and is not security- or performance-sensitive.
=== ticket-maybe-gemini ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: yes
override: n/a
why: The task has cheap uncertainty about whether ticket EXT-4401 covers the change, which is not security- or performance-sensitive, so it proceeds with a final summary note instead of waiting.
=== plan-rive-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: no
why: The engineer explicitly asked for a plan regarding a performance-sensitive animation loading issue that is not fully specified, requiring a brief but no spec since the change is not large.
=== browse-url-gpt ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: yes
override: n/a
why: The Jira URL identifies MCWP-726, but its contents were unavailable, leaving ticket coverage uncertain without evidence warranting a risk stop or another spec.
=== browse-url-gemini ===
triggered: yes
path: proceed
spec: no
stopped: no
jira: yes
override: n/a
why: The ask points to a Jira ticket, which allows the work to proceed without a brief.
=== ticket-maybe-kimi ===
triggered: yes
path: cheap-note
spec: no
stopped: no
jira: yes
override: n/a
why: The engineer explicitly invoked mms-planning, and the request is fully specified except for cheap uncertainty about whether EXT-4401 covers it, with a Jira reference and no security, performance, contract, or large-change trigger.
=== household-mobile-claude ===
triggered: yes
path: spec
spec: yes
stopped: yes
jira: no
override: n/a
why: A new cross-client product area with undecided scope and unknown surfaces counts as large, so it stops for a brief that recommends a spec.
=== household-ext-claude ===
triggered: yes
path: spec
spec: yes
stopped: yes
jira: no
override: no
why: A new product area with undecided scope across Mobile and Extension counts as large and ambiguous, so it stops for a brief that recommends a planning spec, with no Jira reference and no product edits.
=== generic-plan-claude ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: yes
why: The user asked for a plan, and the black screen before an animation is an ambiguous, performance-sensitive render/startup problem that is not a new product area, so it needs a full brief and agreement with no spec.
=== named-skill-claude ===
triggered: no
path: no-trigger
spec: no
stopped: n/a
jira: no
override: yes
why: The engineer named brainstorming and explicitly excluded mms-planning, and the decision order says a skill the engineer named still runs. It also says the generic skill only loses when no other skill was named.
=== household-mobile-gpt ===
triggered: yes
path: spec
spec: yes
stopped: yes
jira: no
override: n/a
why: A new product area spanning Mobile and Extension with undecided scope is large, requiring a brief and spec recommendation before proceeding.
=== household-mobile-kimi ===
triggered: yes
path: full-brief
spec: yes
stopped: yes
jira: no
override: n/a
why: The ask is a new product area with undecided scope, so it is both ambiguous and large: the spec trigger applies, the change stops for a full brief before any edit, and a planning spec is recommended.
=== household-ext-kimi ===
triggered: yes
path: full-brief
spec: yes
stopped: yes
jira: no
override: no
why: The ask designs a new product area (household accounts across Mobile and Extension) with undecided scope and unknown surfaces, which is both ambiguous and large, so it stops for a full brief with a spec recommended before any product edit.
=== household-ext-gpt ===
triggered: yes
path: spec
spec: yes
stopped: yes
jira: no
override: n/a
why: A new product area spanning Mobile and Extension with undecided scope is explicitly large, requiring a brief and spec recommendation before proceeding.
=== generic-plan-kimi ===
triggered: yes
path: short-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The ask names the target, action, and expected result, but fixing a black screen before the onboarding animation is performance-sensitive (startup/animation render path), so step 1 requires stopping with a one-line short brief that names the risk, and no spec because the change is not large.
=== generic-plan-gpt ===
triggered: yes
path: short-brief
spec: no
stopped: yes
jira: no
override: yes
why: The explicit plan request triggers mms-planning over generic planning; fixing the pre-animation black screen is fully specified, performance-sensitive, and not large, requiring a short brief and agreement before edits.
=== named-skill-gpt ===
triggered: yes
path: proceed
spec: no
stopped: no
jira: no
override: no
why: The plan request triggers planning classification, but explicitly named brainstorming is not overridden; this mechanical label change needs no brief or spec, and editing remains deferred as requested.
=== generic-plan-gemini ===
triggered: yes
path: full-brief
spec: no
stopped: yes
jira: no
override: n/a
why: The user explicitly asked for a plan for a fix that is not fully specified, which stops for a brief but does not trigger a large-change spec.
=== household-ext-gemini ===
triggered: yes
path: full-brief
spec: yes
stopped: yes
jira: no
override: n/a
why: The request introduces a large, cross-system new product area with ambiguous scope, requiring a stop for a full brief and a planning spec recommendation.
=== named-skill-kimi ===
triggered: yes
path: no-trigger
spec: no
stopped: n/a
jira: no
override: yes
why: The ask for a plan matches a mms-planning trigger, but the engineer explicitly named brainstorming for the job, so the named skill runs and mms-planning does not win.
=== named-skill-gemini ===
triggered: yes
path: no-trigger
spec: no
stopped: n/a
jira: no
override: yes
why: The user explicitly named the brainstorming skill and excluded mms-planning, which overrides the skill's execution.
=== household-mobile-gemini ===
triggered: yes
path: full-brief
spec: yes
stopped: yes
jira: no
override: n/a
why: The request for a new cross-system product area with undecided scope is large and ambiguous, triggering a full brief and a recommended planning spec before any edits.

Additional Context

Base means the description is in every session. The body still loads only when the skill is selected. No AGENTS.md line.

…a nag

Base installs were skipped while the skills stayed experimental, and planning called jira-brief on every non-revert path. Ship them as stable base skills, and only open Jira when the engineer already brought a ticket.
@mganchas
mganchas merged commit 817f515 into main Nov 17, 2026
29 checks passed
@mganchas
mganchas deleted the fix/planning-base-strict-jira branch November 17, 2026 12:23
@georgewrmarshall georgewrmarshall mentioned this pull request Nov 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants