feat(014): budget-guard skill — implement ADR-003 §G4#19
Open
wendeus0 wants to merge 2 commits into
Open
Conversation
Resolves PENDING_LOG P3: Implementar feature dedicada budget-guard-spec.
Changes:
- core/skills/budget-guard/SKILL.md: new operational skill canonical
(description 176 chars, 4 triggers, gate per feature, circuit breaker
>5%, Haiku default, override via CLAUDE.md local, stateless cross-feature)
- 3 domain gates: BUDGET_OK, BUDGET_WARN, BUDGET_BLOCKED declared
formally in `# Estados de saída`
- 5 meta skills restored Task invocation:
- fix-feature, implement-feature, refactor-feature, technical-triage,
test-architecture-plan: prose neutra → `budget-guard` via `Task`
- ADR-003: G4 marked active (was PENDING-SPEC); distribution 1A → 2A
- PENDING_LOG: P3 entry marked DONE
- memory/next_steps.md: line 2 marked done
Validation:
- audit.py (full): 0 findings — REFS no longer reports PHANTOM
- validate-spec --feature 014-budget-guard: 0 findings
- smoke 9/9
- sync-skills drift 0 (18 skills propagated to 3 mirrors)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
CodeAnt AI is reviewing your PR. |
📝 WalkthroughWalkthroughThis PR implements the G4 "budget-guard" feature: adds a new canonical skill that emits budget gates and model recommendations, updates five meta-skills to invoke it via Task instead of ADR-003 prose, and marks the pending feature as DONE with ADR-003 §G4 activated and zero audit findings. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
actor Developer
participant Orchestrator
participant BudgetGuard as "budget-guard (Skill)"
participant ModelSelector
participant Runtime
Developer->>Orchestrator: submit task with triage_class, feature_scope, current_model, running_cost_pct?
Orchestrator->>BudgetGuard: TaskInvoke(inputs)
BudgetGuard-->>Orchestrator: BUDGET_OK / BUDGET_WARN / BUDGET_BLOCKED + model rec
Orchestrator->>ModelSelector: choose worker/model based on gate & triage
ModelSelector->>Runtime: execute using selected model/worker
Runtime-->>Orchestrator: result / telemetry (or telemetry unavailable)
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~22 minutes Possibly related PRs
Suggested labels
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
codeant-ai
Bot
added
the
size:L
|
CodeAnt AI finished reviewing your PR. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 6
🧹 Nitpick comments (1)
core/skills/test-architecture-plan/SKILL.md (1)
66-68: ⚡ Quick winAlign trigger wording with the canonical triage classes.
Line 67 uses “projetos grandes”, while the feature contract standardizes triggers as
high,multi-step,security,large-context. Using the same taxonomy here avoids ambiguity across meta-skills.Suggested patch
-2. **Controle de custo:** projetos grandes — invoque a skill `budget-guard` via `Task` antes de iniciar o mapeamento completo. +2. **Controle de custo:** se o `triage` indicar `high`, `multi-step`, `security` ou `large-context`, invoque a skill `budget-guard` via `Task` antes de iniciar o mapeamento completo.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@core/skills/test-architecture-plan/SKILL.md` around lines 66 - 68, The phrase "projetos grandes" in the "Controle de custo" step is not using the canonical trigger taxonomy; update that wording to use the standardized trigger token `large-context` (and ensure the doc references the canonical trigger set: `high`, `multi-step`, `security`, `large-context`) so the line reads something like "Controle de custo: `large-context` — invoque a skill `budget-guard` via `Task`..."; apply the same canonical trigger wording anywhere else in SKILL.md that uses informal trigger phrases to avoid ambiguity across meta-skills.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@core/skills/budget-guard/SKILL.md`:
- Around line 55-57: The fenced output-format block in SKILL.md currently lacks
a language tag; update the triple-backtick fence that contains
"<BUDGET_OK|BUDGET_WARN|BUDGET_BLOCKED>: <recomendação curta com modelo
sugerido>" to include a language identifier (e.g., add "text" after the opening
```), so the block becomes ```text ... ``` to satisfy markdownlint MD040 (refer
to the fenced block showing the budget status tokens).
- Line 31: The doc has inconsistent threshold wording for the circuit breaker:
the phrase "Circuit breaker auto-disable em >5%" conflicts with the later lines
that say "5% and above"; pick one policy (either strictly greater than 5% or
greater-or-equal to 5%) and make the text consistent everywhere—update the
header phrase and any occurrences referring to the threshold and the behavior
that returns BUDGET_BLOCKED so they all use the same operator (e.g., change
">5%" to ">=5%" or vice versa) and ensure the description of required
override/flow control remains aligned with that chosen semantics.
In `@features/014-budget-guard/PLAN.md`:
- Around line 13-19: The fenced code block containing the diagram (starting with
implement-feature ──▶ Task(budget-guard) ──▶ skill avalia contexto ──▶ retorna
BUDGET_{OK,WARN,BLOCKED}) is missing a language identifier; add the language tag
`text` to the opening triple backticks so the block becomes ```text to satisfy
MD040 and keep linting green while preserving the diagram content.
In `@features/014-budget-guard/TASKS.md`:
- Around line 19-24: The fenced code block that contains the critical-path
diagram starting with "T1 (criar skill) ──▶ T2 (restaurar 5 invocações)" is
missing a language tag; add "text" after the opening triple backticks so the
block reads ```text and keep the block content unchanged to satisfy MD040
linting.
In `@memory/next_steps.md`:
- Line 8: Update the document header date after the new completion entry: locate
the "Atualizado: 2026-04-28" header in next_steps.md and change it to
"Atualizado: 2026-04-30" to match the entry "Concluído em 2026-04-30 via feature
`014-budget-guard`" so the header timestamp is consistent with the recorded
completion.
In `@PENDING_LOG.md`:
- Line 34: Update the DONE entry in PENDING_LOG.md to correct the output-state
count: change "5 estados de saída declarados" to "3 estados de saída declarados
(BUDGET_OK, BUDGET_WARN, BUDGET_BLOCKED)" to match the actual states declared in
core/skills/budget-guard/SKILL.md; ensure the rest of the line (feature name,
triggers, gate, circuit breaker, model, overrides, Task invocation, ADR-003,
auditor note) remains unchanged.
---
Nitpick comments:
In `@core/skills/test-architecture-plan/SKILL.md`:
- Around line 66-68: The phrase "projetos grandes" in the "Controle de custo"
step is not using the canonical trigger taxonomy; update that wording to use the
standardized trigger token `large-context` (and ensure the doc references the
canonical trigger set: `high`, `multi-step`, `security`, `large-context`) so the
line reads something like "Controle de custo: `large-context` — invoque a skill
`budget-guard` via `Task`..."; apply the same canonical trigger wording anywhere
else in SKILL.md that uses informal trigger phrases to avoid ambiguity across
meta-skills.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 06b4dd1a-7d7f-4b8b-a6e5-4414cd0668cd
📒 Files selected for processing (14)
PENDING_LOG.mdcore/skills/budget-guard/SKILL.mdcore/skills/fix-feature/SKILL.mdcore/skills/implement-feature/SKILL.mdcore/skills/refactor-feature/SKILL.mdcore/skills/technical-triage/SKILL.mdcore/skills/test-architecture-plan/SKILL.mddocs/adr/ADR-003-ghost-skills-treatment.mdfeatures/014-budget-guard/CONTRACT.mdfeatures/014-budget-guard/PLAN.mdfeatures/014-budget-guard/REPORT.mdfeatures/014-budget-guard/SPEC.mdfeatures/014-budget-guard/TASKS.mdmemory/next_steps.md
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
2 issues found across 14 files
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="core/skills/budget-guard/SKILL.md">
<violation number="1" location="core/skills/budget-guard/SKILL.md:3">
P2: Use a single breaker threshold definition (`>= 5%` or `> 5%`) across the whole skill; current mixed wording creates ambiguous behavior at exactly 5%.</violation>
</file>
<file name="features/014-budget-guard/TASKS.md">
<violation number="1" location="features/014-budget-guard/TASKS.md:29">
P2: O comando de validação de T2 usa o padrão com `\`` e não encontra as ocorrências reais, quebrando o critério de saída documentado.</violation>
</file>
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
- Threshold uniforme ≥5% em SKILL.md (description, regra, processo) e em PENDING_LOG. Antes ambíguo: descrição dizia >5% mas processo e tabela usavam ≥5%. - Adicionado language tag `text` aos 3 fenced blocks sem MD040 (SKILL.md output format, PLAN.md diagrama, TASKS.md critical path). - TASKS.md T2: comando grep simplificado para um loop com -qF e testado funcional (saída vazia confirma 5/5 invocações restauradas). - memory/next_steps.md header sincronizado para 2026-04-30. - PENDING_LOG: contagem corrigida para 3 estados de saída (BUDGET_OK/ WARN/BLOCKED), não 5 como estava. - audit.py EXTERNAL_TOOLS: adicionados model IDs Anthropic (claude- haiku-4-5, sonnet-4-5/4-6, opus-4-6/4-7) e classes textuais (multi- step, large-context) para evitar PHANTOM REFS em refs literais. Validation: - audit.py (full): 0 findings - smoke 9/9 - sync-skills drift 0 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@core/skills/budget-guard/SKILL.md`:
- Around line 31-36: The documentation currently leaves `BUDGET_BLOCKED`
ambiguous between an informational flag and a mandatory stop; update SKILL.md to
state that `BUDGET_BLOCKED` is an enforceable circuit-breaker which requires the
runtime invoker to abort execution unless an explicit override is present, and
describe the override mechanism (`budget_guard_model` and
`budget_guard_threshold_pct` in CLAUDE.md) and how to record an explicit
override (e.g., runtime must set an explicit "override" token or reduce scope)
before continuing; keep the recommended default model `claude-haiku-4-5`, the 5%
threshold language, and note that the skill still remains stateless but the
invoker must enforce the abort behavior when `BUDGET_BLOCKED` is returned.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 78dd685b-97cd-4c81-835f-b58988dc486f
📒 Files selected for processing (6)
PENDING_LOG.mdcore/skills/budget-guard/SKILL.mdfeatures/009-skill-harmonization/scripts/audit.pyfeatures/014-budget-guard/PLAN.mdfeatures/014-budget-guard/TASKS.mdmemory/next_steps.md
🚧 Files skipped from review as they are similar to previous changes (1)
- PENDING_LOG.md
Comment on lines
+31
to
+36
| - **Circuit breaker auto-disable em ≥5%** — se o custo acumulado atingir ou superar 5% do orçamento estimado da feature, retorne `BUDGET_BLOCKED`; runtime invocador exige override explícito ou redução de escopo antes de prosseguir. | ||
| - **Modelo recomendado default: Haiku (`claude-haiku-4-5`)** — função classificatória/heurística simples; não exige premium. | ||
| - **Override por projeto** registrável em `CLAUDE.md` local com `budget_guard_model: <model-id>` ou `budget_guard_threshold_pct: <N>` — projetos com perfis distintos podem afrouxar/apertar o gate. | ||
| - **Stateless cross-feature** — não persiste em arquivo; estado vive na sessão da feature ativa. | ||
| - **Veredito é informativo, não bloqueante automático** — `BUDGET_BLOCKED` exige que o runtime invocador respeite, mas a skill em si não interrompe execução. | ||
|
|
There was a problem hiding this comment.
BUDGET_BLOCKED ficou sem obrigação operacional explícita para o invocador.
Line 31 descreve auto-disable, mas Line 35 permite interpretação “apenas informativa”. Isso enfraquece o gate e pode deixar o bloqueio opcional nas skills consumidoras.
🔒 Patch sugerido para fechar a ambiguidade
-- **Veredito é informativo, não bloqueante automático** — `BUDGET_BLOCKED` exige que o runtime invocador respeite, mas a skill em si não interrompe execução.
+- **Veredito é gate obrigatório para o invocador** — ao receber `BUDGET_BLOCKED`, o runtime deve interromper escalonamento premium e só prosseguir com override explícito (ou redução de escopo) registrado.🧰 Tools
🪛 LanguageTool
[misspelling] ~31-~31: Esta é uma palavra só.
Context: ...pct` cruzar limiar. - Circuit breaker auto-disable em ≥5% — se o custo acumulado atingir...
(AUTO)
[misspelling] ~33-~33: Possível erro ortográfico.
Context: ...** registrável em CLAUDE.md local com budget_guard_model: <model-id> ou budget_guard_threshold_pct: <N> ...
(PT_MULTITOKEN_SPELLING_HYPHEN)
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@core/skills/budget-guard/SKILL.md` around lines 31 - 36, The documentation
currently leaves `BUDGET_BLOCKED` ambiguous between an informational flag and a
mandatory stop; update SKILL.md to state that `BUDGET_BLOCKED` is an enforceable
circuit-breaker which requires the runtime invoker to abort execution unless an
explicit override is present, and describe the override mechanism
(`budget_guard_model` and `budget_guard_threshold_pct` in CLAUDE.md) and how to
record an explicit override (e.g., runtime must set an explicit "override" token
or reduce scope) before continuing; keep the recommended default model
`claude-haiku-4-5`, the 5% threshold language, and note that the skill still
remains stateless but the invoker must enforce the abort behavior when
`BUDGET_BLOCKED` is returned.
|
CodeAnt AI is running the review. |
codeant-ai
Bot
added
size:L
Sequence DiagramThis PR introduces the budget-guard skill and reconnects five meta skills so they route high-impact work through a per-feature cost gate before choosing an LLM model. sequenceDiagram
participant User
participant Meta as Meta skills
participant Triage as Triage skill
participant Task as Task runtime
participant BudgetGuard as Budget guard skill
User->>Meta: Request complex feature work
Meta->>Triage: Run triage to classify impact and triggers
Triage-->>Meta: Return triage class and triggers
Meta->>Task: Invoke budget guard with scope, model, running cost
Task->>BudgetGuard: Evaluate cost gate for this feature
BudgetGuard-->>Task: Return budget state and model recommendation
Task-->>Meta: Deliver BUDGET_OK or WARN or BLOCKED decision
Meta-->>User: Continue with chosen model or require override based on gate
Generated by CodeAnt AI |
|
CodeAnt AI finished running the review. |
|
CodeAnt AI is running the review. |
codeant-ai
Bot
added
size:L
Sequence DiagramThis PR introduces the budget-guard skill and wires five meta skills to invoke it via Task when triage flags high cost risk, so it returns a budget verdict that guides model choice and can block premium usage once a cost threshold is reached. sequenceDiagram
participant MetaSkill
participant Triage
participant Runtime
participant BudgetGuard
MetaSkill->>Triage: Classify task risk and scope
Triage-->>MetaSkill: triage_class and context
MetaSkill->>Runtime: Task call to budget-guard with cost data
Runtime->>BudgetGuard: Invoke skill with triage_class and running_cost_pct
BudgetGuard->>BudgetGuard: Compute budget verdict
BudgetGuard-->>Runtime: BUDGET_OK or WARN or BLOCKED with model advice
Runtime-->>MetaSkill: Return budget verdict
alt Budget ok or warn
MetaSkill->>Runtime: Proceed with recommended model
else Budget blocked
MetaSkill->>Runtime: Require override or reduce scope before premium
end
Generated by CodeAnt AI |
![codeant-ai[bot]](https://avatars.githubusercontent.com/in/646884?s=60&v=4){kind=small}
Comment on lines
+49
to
+50
| "claude-haiku-4-5", "claude-sonnet-4-5", "claude-sonnet-4-6", | ||
| "claude-opus-4-6", "claude-opus-4-7", |
There was a problem hiding this comment.
Suggestion: Model identifiers were added to EXTERNAL_TOOLS, but they are not executable tools. Because classify() treats EXTERNAL as valid in via Task/invoque checks, a mistaken invocation like claude-opus-4-7 via Task will be silently accepted instead of flagged as PHANTOM. Keep model IDs out of EXTERNAL_TOOLS and handle them in a separate non-invocation whitelist if you only want to suppress prose/backtick false positives. [api mismatch]
Severity Level: Major ⚠️
- ❌ REFS detector skips phantom model Task invocations in skills.
- ⚠️ Skill harmonization audit misses miswired model invocation references.Steps of Reproduction ✅
1. Open `core/skills/budget-guard/SKILL.md` (lines 10–18 show the `# Quando usar` section
with triggers like `high`, `multi-step`, etc.) and add a new bullet such as `- invoke
\`claude-opus-4-7\` via \`Task\`` to simulate a mis-specified Task target in a real skill
document.
2. From the repository root, run `python3
features/009-skill-harmonization/scripts/audit.py --detector REFS`; this executes `run()`
in `features/009-skill-harmonization/scripts/audit.py:70-107`, which loads all skills,
constructs a `Catalog`, and calls `detect_refs()` for each skill (lines 368-392).
3. In `detect_refs()` (lines 368-399 of
`features/009-skill-harmonization/scripts/audit.py`), the `_VIA_TASK_RE` regex defined at
lines 339-340 matches the new `claude-opus-4-7 via Task` text, and the detector invokes
`kind, mapped = catalog.classify(token)` at line 380.
4. Because `EXTERNAL_TOOLS` includes `"claude-opus-4-7"` (lines 34-52, specifically
49-50), `Catalog.classify()` at lines 291-301 returns `("EXTERNAL", "claude-opus-4-7")`,
so the `if kind in ("SKILL", "SUBAGENT", "NATIVE", "EXTERNAL", "ALIAS", "MCP", "PLUGIN"):`
check in `detect_refs()` (line 399-400) treats this as a valid invocation and skips
emitting a `PHANTOM ... via Task` finding, even though `claude-opus-4-7` is a model
identifier, not a Task-invocable skill or tool, confirming that misclassified model IDs
bypass phantom-reference detection.Fix in Cursor | Fix in VSCode Claude
(Use Cmd/Ctrl + Click for best experience)
Prompt for AI Agent 🤖
This is a comment left during a code review.
**Path:** features/009-skill-harmonization/scripts/audit.py
**Line:** 49:50
**Comment:**
*Api Mismatch: Model identifiers were added to `EXTERNAL_TOOLS`, but they are not executable tools. Because `classify()` treats `EXTERNAL` as valid in `via Task`/`invoque` checks, a mistaken invocation like `claude-opus-4-7` via `Task` will be silently accepted instead of flagged as PHANTOM. Keep model IDs out of `EXTERNAL_TOOLS` and handle them in a separate non-invocation whitelist if you only want to suppress prose/backtick false positives.
Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
Once fix is implemented, also check other comments on the same PR, and ask user if the user wants to fix the rest of the comments as well. if said yes, then fetch all the comments validate the correctness and implement a minimal fix| "spring-cloud-contract-stub-runner", | ||
| "claude-haiku-4-5", "claude-sonnet-4-5", "claude-sonnet-4-6", | ||
| "claude-opus-4-6", "claude-opus-4-7", | ||
| "multi-step", "large-context", |
There was a problem hiding this comment.
Suggestion: Trigger labels were added as external tools, which broadens the validator in the wrong dimension: multi-step/large-context will now be treated as legitimate callable tokens in chain and invocation detectors. This creates false negatives where malformed workflow steps are no longer reported. Keep these as contextual trigger terms (not tools), or whitelist them only in prose-specific checks. [logic error]
Severity Level: Major ⚠️
- ❌ CHAINS detector treats trigger labels as callable workflow steps.
- ⚠️ Workflow chain validation loses coverage for phantom tokens.Steps of Reproduction ✅
1. In `core/skills/budget-guard/SKILL.md` (lines 37-59 define `# Processo`), add a
chain-style line such as `multi-step → budget-guard` or `[multi-step] → [budget-guard]` to
describe a workflow step using the `multi-step` trigger label in chain notation.
2. Run `python3 features/009-skill-harmonization/scripts/audit.py --detector CHAINS` from
the repository root; this calls `run()` in
`features/009-skill-harmonization/scripts/audit.py:70-107`, which in turn invokes
`detect_chains()` for each skill via the `PER_SKILL_DETECTORS` registry at lines 55-63.
3. Inside `detect_chains()` (originally around lines 415-45 of
`features/009-skill-harmonization/scripts/audit.py`), `_CHAIN_LINE_RE` defined at lines
~411-412 matches the new arrow line, the loop extracts `multi-step` as a token (lines
~23-31 in that function), and `kind, mapped = catalog.classify(token)` at line ~436 calls
`Catalog.classify()` with `"multi-step"`.
4. Because `EXTERNAL_TOOLS` includes `"multi-step"` (lines 34-52, specifically line 51),
`Catalog.classify()` (lines 291-301) returns `("EXTERNAL", "multi-step")`, so
`detect_chains()`'s `if kind in ("SKILL", "SUBAGENT", "ALIAS", "NATIVE", "EXTERNAL",
"MCP", "PLUGIN"):` check at line ~437 treats `multi-step` as a legitimate callable token
and skips emitting a `PHANTOM step` finding; this confirms that adding trigger labels like
`multi-step`/`large-context` to `EXTERNAL_TOOLS` suppresses CHAINS phantom-token
validation for malformed workflow steps.Fix in Cursor | Fix in VSCode Claude
(Use Cmd/Ctrl + Click for best experience)
Prompt for AI Agent 🤖
This is a comment left during a code review.
**Path:** features/009-skill-harmonization/scripts/audit.py
**Line:** 51:51
**Comment:**
*Logic Error: Trigger labels were added as external tools, which broadens the validator in the wrong dimension: `multi-step`/`large-context` will now be treated as legitimate callable tokens in chain and invocation detectors. This creates false negatives where malformed workflow steps are no longer reported. Keep these as contextual trigger terms (not tools), or whitelist them only in prose-specific checks.
Validate the correctness of the flagged issue. If correct, How can I resolve this? If you propose a fix, implement it and please make it concise.
Once fix is implemented, also check other comments on the same PR, and ask user if the user wants to fix the rest of the comments as well. if said yes, then fetch all the comments validate the correctness and implement a minimal fix
|
CodeAnt AI finished running the review. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
User description
Summary
Resolve P3 do PENDING_LOG implementando ADR-003 §G4 (council 2026-04-27 adopted): cria a skill canônica
budget-guarde restaura as 5 invocações nas skills meta.claude-haiku-4-5-20251001triage)features/NNN/budget_guard_model/budget_guard_threshold_pctem CLAUDE.md localMudanças
core/skills/budget-guard/SKILL.mdcore/skills/{fix-feature,implement-feature,refactor-feature,technical-triage,test-architecture-plan}/SKILL.md\budget-guard` via `Task``docs/adr/ADR-003-ghost-skills-treatment.mdPENDING_LOG.mdmemory/next_steps.mdfeatures/014-budget-guard/{SPEC,PLAN,TASKS,CONTRACT,REPORT}.mdValidação
bash scripts/validate-spec.sh --feature 014-budget-guard→ 0 findingspython3 features/009-skill-harmonization/scripts/audit.py(full) → 0 findingsaudit.py --detector REFS→ 0 (token deixa de ser PHANTOM)audit.py --detector SKILL_FORMAT→ 0audit.py --detector CONTRACT→ 0 (gatesBUDGET_*declarados formalmente)bash tests/smoke/all.sh→ 9/9 verdesbash tools/sync-skills.sh --check→ drift 0 (18 skills propagadas para 3 mirrors)grep -l 'budget-guard\via `Task' core/skills/{...}/SKILL.md` → 5/5Decisões de execução
test-design— PLAN.md confirma 0 condições binárias (skill é definição de contrato textual). Camada unitária via detectores existentes.code-reviewSKIP — diff doc-only (1 SKILL nova + 5 edições linha-única + 3 docs). Sem lógica nova testável.security-reviewSKIP — superfície de ataque zero (apenas markdown).Test plan
🤖 Generated with Claude Code
CodeAnt-AI Description
Add the canonical
budget-guardskill and restore its use in meta skillsWhat Changed
budget-guardskill that tells meta skills when to keep, lower, or stop using a higher-cost model based on the current feature’s budgetbudget-guardcalls in the five meta skills that use cost control so they invoke the skill again instead of only mentioning the ADRImpact
✅ Clearer cost-gating for large or risky tasks✅ Fewer missed budget checks in feature workflows✅ No more phantom references for budget-guard🔄 Retrigger CodeAnt AI Review
Details
💡 Usage Guide
Checking Your Pull Request
Every time you make a pull request, our system automatically looks through it. We check for security issues, mistakes in how you're setting up your infrastructure, and common code problems. We do this to make sure your changes are solid and won't cause any trouble later.
Talking to CodeAnt AI
Got a question or need a hand with something in your pull request? You can easily get in touch with CodeAnt AI right here. Just type the following in a comment on your pull request, and replace "Your question here" with whatever you want to ask:
This lets you have a chat with CodeAnt AI about your pull request, making it easier to understand and improve your code.
Example
Preserve Org Learnings with CodeAnt
You can record team preferences so CodeAnt AI applies them in future reviews. Reply directly to the specific CodeAnt AI suggestion (in the same thread) and replace "Your feedback here" with your input:
This helps CodeAnt AI learn and adapt to your team's coding style and standards.
Example
Retrigger review
Ask CodeAnt AI to review the PR again, by typing:
Check Your Repository Health
To analyze the health of your code repository, visit our dashboard at https://app.codeant.ai. This tool helps you identify potential issues and areas for improvement in your codebase, ensuring your repository maintains high standards of code health.