bug(doctor): "Deployed Skill Count" check ignores skillListingBudgetFraction override; warns persistently after remediation

[ChronodeAi]

TL;DR

aiwg doctor has two checks that measure the same thing (deployed skill listing tokens) against different baselines. The "Skill Budget" check correctly reads skillListingBudgetFraction from ~/.claude/settings.json and flips to ✓ when the user raises the budget. The "Deployed Skill Count" check is hard-coded to compare against the vendor default (500 tokens) and continues to warn even after the user has taken the documented remediation. Result: doctor produces a persistent, redundant warning that can't be cleared without removing skills.

Environment (REQUIRED — issues missing any field will be returned for clarification)

• AIWG version: 2026.5.12 [stable]
• Operating system: Darwin mac-studio.tail10294f.ts.net 24.6.0 Darwin Kernel Version 24.6.0: Mon Jan 19 22:01:58 PST 2026; root:xnu-11417.140.69.708.3~1/RELEASE_ARM64_T6041 arm64
• Provider (the AIWG harness you were running): claude-code
• Node version: v26.0.0
• Install location: npm-global

Steps to reproduce

# 1. Deploy enough skills to exceed the 500-token default
aiwg use all --provider claude

# 2. Verify doctor warns about both checks
aiwg doctor 2>&1 | grep -E "Skill Budget|Deployed Skill Count"
# ⚠ Claude Code Skill Budget: EXCEEDS DEFAULT (2.58×) — ...
# ⚠ Claude Code Deployed Skill Count: 28 deployed skills estimate 1,289 tokens, above ...

# 3. Apply the documented remediation: raise the budget fraction
# (edit ~/.claude/settings.json to add "skillListingBudgetFraction": 0.03)

# 4. Re-run doctor
aiwg doctor 2>&1 | grep -E "Skill Budget|Deployed Skill Count"

Expected behavior

Both checks should respect the user's skillListingBudgetFraction override. After raising the budget to cover deployed skills, both should report ✓.

Actual behavior

Only the "Skill Budget" check flips to ✓. The "Deployed Skill Count" check continues to warn against the vendor default:

✓ Claude Code Skill Budget: tight (0.86×) — 28 skills (avg 165 chars desc), est. 1,289 tokens vs 1,500 tokens budget — 3.00% × 200,000 ctx (override in ~/.claude/settings.json)
⚠ Claude Code Deployed Skill Count: 28 deployed skills estimate 1,289 tokens, above Claude Code's default listing budget (500 tokens). Run `aiwg use all` for workspace-aware filtering or `aiwg list --deployed` to inspect include/exclude reasons.

The two checks now contradict each other: one says "your configured budget is fine", the other says "you're above the default" — without acknowledging the user has explicitly opted out of the default.

Suggested fix

tools/cli/doctor.mjs line 246-253 (the "Deployed Skill Count" check) is hard-coded to use defaultBudgetTokens. The "Skill Budget" check (around line 321-338) already reads the configured fraction and computes the effective budget. The Deployed Skill Count check should use the same effective budget:

// Current (line 246-253):
if (provName === 'claude') {
  const defaultBudgetTokens = Math.floor(
    (CLAUDE_DEFAULT_CONTEXT_WINDOW * CLAUDE_DEFAULT_BUDGET_FRACTION) / CHARS_PER_TOKEN,
  );
  if (stats.totalTokens > defaultBudgetTokens) {
    check(`${label} Deployed Skill Count`, 'warn', `... above Claude Code's default listing budget ...`);
  }
}

// Suggested: read the configured fraction (same code path as the Skill Budget check)
const fraction = readSkillListingBudgetFraction() ?? CLAUDE_DEFAULT_BUDGET_FRACTION;
const effectiveBudgetTokens = Math.floor(
  (CLAUDE_DEFAULT_CONTEXT_WINDOW * fraction) / CHARS_PER_TOKEN,
);
if (stats.totalTokens > effectiveBudgetTokens) {
  check(`${label} Deployed Skill Count`, 'warn', `... above your configured listing budget (${effectiveBudgetTokens} tokens) ...`);
}

Alternatively, if the intent of "Deployed Skill Count" is specifically to surface "you're above vendor defaults" as an informational note, drop the redundant warn severity to note/info-level (the ○ marker doctor already uses for optional things) so it doesn't count against the warning total.

The first option is cleaner — having two warn-severity checks for the same condition that disagree on their baseline is confusing.

Acceptance criteria

• After setting skillListingBudgetFraction to a value that covers the deployed skill tokens, both the "Skill Budget" and "Deployed Skill Count" checks pass.
• Doctor warning count is consistent — no redundant warnings for the same condition once the user has applied the documented remediation.
• Documentation for aiwg doctor describes whether "Deployed Skill Count" reads the override or stays pinned to defaults (so behavior is unambiguous either way).

Related

• Issues: none found via gh issue list -R jmagly/aiwg --search "Deployed Skill Count budget override" or --search "skillListingBudgetFraction"
• PRs:
• Source: tools/cli/doctor.mjs:246-253 (Deployed Skill Count check) vs tools/cli/doctor.mjs:321-338 (Skill Budget check that does read the override)
• Related: bug(cli): aiwg remove --provider <name> crashes with TypeError instead of supporting or rejecting the flag #118 (recent aiwg remove --provider TypeError, also filed against aiwg doctor companion CLI surface)