Configure Codacy

Configure Codacy

Glossary: See glossary.md for shared definitions of Codacy concepts (issues, findings, severity, coverage, tools, patterns, etc.).

This skill tailors Codacy configuration to a project's actual stack and coding conventions. It discovers the repository's languages and frameworks, initializes a broad set of tools and patterns, runs analysis, then intelligently cuts noise — producing a clean, high-signal configuration with a full audit trail of what changed and why.

Prerequisites

• Codacy Analysis CLI (codacy-analysis) with discover and init --auto support. If the CLI does not support these commands, update it running: npm i -g @codacy/analysis-cli. See codacy-analysis-cli for setup.
• Codacy Cloud CLI (codacy) — needed to know if the repo is on Codacy Cloud and if so, fetch issues from the cloud-only tools. See codacy-cloud-cli for setup.

Both CLIs share credentials at ~/.codacy/credentials, so a single login covers both.

How configuration works

All configuration is done locally via .codacy/codacy.config.json. Edit the file, run analysis, see results instantly — no push or cloud reanalysis needed. Once tuned, the configuration can be imported to Codacy Cloud in one step.

The key principle: start broad, then cut noise using data. Initialize with maximum pattern coverage via init --auto, run analysis to see the full issue landscape, then use the severity/category distribution to decide what to disable or tune.

Read the config format reference for the full schema before editing — field names matter (e.g., the exclusion field is exclude, not excludePaths). To disable a pattern, remove it from the patterns array. To disable a tool, remove the entire tool entry.

Organization standards take precedence. If a pattern is enforced by a Coding Standard at the organization level, it cannot be changed at the repository level.

Local config only. The .codacy/codacy.config.json file is used exclusively for local analysis. Committing or pushing it to the repository has NO effect on Codacy Cloud analysis. To apply the configuration to Codacy Cloud, use the import command (codacy tools ... --import).

Invocation modes

The skill supports three modes that control whether configuration is imported to Codacy Cloud after tuning:

• Interactive (default): After tuning and presenting results, prompt the user via AskUserQuestion whether to import the configuration to Codacy Cloud. This is the default when no arguments are provided.
• Auto-import: If the user's invocation arguments contain the word "import" (e.g., /configure-codacy import configuration, /configure-codacy --import), skip the import prompt and import automatically after presenting results. Useful for CI runs and automation.
• Local-only: If the repo is not on Codacy Cloud (determined in Step 0), never ask about importing. Present results and note that the config is ready for local analysis only.

Additionally, the force flag controls Coding Standard handling during import:

• Force disabled (default): If the import encounters Coding Standard conflicts (409 errors), do NOT automatically retry with --force. Instead, present the conflicts to the user via AskUserQuestion and let them decide whether to unlink the Coding Standards.
• Force enabled: If the user's invocation arguments contain the word "force" (e.g., /configure-codacy import force), automatically retry with --force when Coding Standard conflicts occur.

Parse the invocation arguments at the very start. Set internal flags that Step 6 will reference:

• If args contain "import" → auto-import mode
• If args contain "force" → force mode (allows automatic Coding Standard unlinking)
• Else → interactive mode (may become local-only if Step 0 finds the repo is not on Cloud)

Tailored configuration workflow

Configuration Progress:
- [ ] Step 0: Check environment and capture baseline
- [ ] Step 1: Discover repository stack
- [ ] Step 2: Initialize auto-tuned configuration
- [ ] Step 3: Run broad-config baseline analysis
- [ ] Step 4: Smart noise evaluation and tuning
- [ ] Step 5: Show local results
- [ ] Step 6: Cloud import (conditional)
- [ ] Step 7: Clean-up

Step 0: Check environment and capture baseline

This step determines the starting point and captures the BEFORE metrics for the final summary. The flow depends on two conditions: whether the repo is on Codacy Cloud, and whether a local config already exists.

0a. Create temp directory and check Codacy Cloud status

Create the temporary directory for intermediate analysis files:

mkdir -p .codacy/tmp

The Cloud CLI auto-detects provider, organization, and repository from the git remote origin URL — no need to derive these manually. All codacy commands below can be run without explicit <provider> <org> <repo> when inside the repo. For clarity, the examples use the explicit form; replace with auto-detected short form in practice.

CLI output caveat: Both codacy and codacy-analysis CLIs write progress lines (e.g., - Fetching tools..., - Fetching issues...) to stderr before the actual JSON output. When piping CLI output directly to jq, redirect stderr to avoid parse errors:

codacy repository --output json 2>/dev/null | jq '...'

Check Cloud status:

codacy repository --output json

If this succeeds, the repo is on Codacy Cloud. Also list enabled tools to identify cloud-only tools later:

codacy tools --output json 2>/dev/null | jq '[.[] | select(.settings.isEnabled == true) | {name, isClientSide}]'

If it fails (repo not on Codacy, no auth, or no Cloud CLI), note that cloud features will be skipped — the local workflow is fully self-contained. Set the invocation mode to local-only (see "Invocation modes" section).

0b. Capture baseline (BEFORE reference)

Branch based on Cloud status and local config existence. The baseline captured here is the true BEFORE reference for the summary in Step 5.

Track A — Repo is on Codacy Cloud (authoritative baseline):

The Cloud configuration is the authoritative source. Use init --remote to fetch it.

Important: init --remote will fail if .codacy/codacy.config.json already exists. Always delete it first (the backup copy step happens before the delete):

# Delete any existing config first — init --remote refuses to overwrite
rm -f .codacy/codacy.config.json

# Fetch the current Cloud configuration
codacy-analysis init --remote <provider> <org> <repo>

# Store a copy for later merging and comparison
cp .codacy/codacy.config.json .codacy/tmp/codacy-remote-config.json

# Record BEFORE metrics
jq '[.tools[].patterns | length] | add' .codacy/codacy.config.json
jq '.tools | length' .codacy/codacy.config.json

# Run analysis with the Cloud config to capture the BEFORE issue landscape
codacy-analysis analyze --install-dependencies --output-format json --output .codacy/tmp/codacy-remote-results.json

# Record BEFORE issue count and runtime
jq '.issues | length' .codacy/tmp/codacy-remote-results.json
jq '.metadata.durationMs' .codacy/tmp/codacy-remote-results.json

Fetch the Cloud issue overview — this gives per-pattern issue counts, false positive counts, and suggested actions across the entire repository without downloading every individual issue:

codacy issues -O -o json > .codacy/tmp/codacy-cloud-overview.json

This overview data is valuable for tuning decisions in Step 4: it shows which patterns produce the most issues on Cloud (the production truth), includes false positive rates per pattern, and generates suggested actions to reduce noise (identifying patterns accounting for 10%+ of all issues or 3x the average). The suggested actions include ready-to-run disable commands, adapted for coding standard or config-file constraints. Save it for later comparison.

Also check for cloud-only tools — tools enabled in Cloud but not available in the local Analysis CLI. Get the local CLI's supported tools via codacy-analysis info, then compare against the Cloud-enabled tools list above. Any Cloud-enabled tool not in the info output is cloud-only (e.g., SonarSharp, Codacy ScalaMeta Pro). If cloud-only tools exist and have issues, fetch them:

codacy issues --output json > .codacy/tmp/codacy-remote-cloud-results.json

Track B — NOT on Cloud, but local .codacy/codacy.config.json exists:

# Store a copy for later merging and comparison
cp .codacy/codacy.config.json .codacy/tmp/codacy-previous-config.json

# Record BEFORE metrics
jq '[.tools[].patterns | length] | add' .codacy/codacy.config.json
jq '.tools | length' .codacy/codacy.config.json

# Run analysis with the existing local config
codacy-analysis analyze --install-dependencies --output-format json --output .codacy/tmp/codacy-previous-results.json

# Record BEFORE issue count and runtime
jq '.issues | length' .codacy/tmp/codacy-previous-results.json
jq '.metadata.durationMs' .codacy/tmp/codacy-previous-results.json

Track C — No Cloud, no local config:

Record BEFORE as unconfigured: 0 patterns, 0 tools, 0 issues, null runtime. No analysis to run.

Save the BEFORE metrics from whichever track ran. The broad config metrics from Steps 2-3 are internal only — they are NOT reported to the user.

0c. Cloud noise pre-evaluation (Track A only)

If the Cloud overview was fetched (.codacy/tmp/codacy-cloud-overview.json exists), analyze it now to identify patterns that are already known to be noisy based on Cloud production data. This saves time by pre-disabling known noise before running local analysis.

Start by checking the overview's suggested actions — the CLI already identifies patterns accounting for 10%+ of all issues or 3x the average per-pattern count and provides ready-to-run disable commands. Use these suggestions as the primary signal, then supplement with the criteria below.

Parse the overview's patterns array (each entry has {id, title, total}) and false positive counts:

# Issue counts per pattern
jq '[.overview.patterns[] | {id, title, total}] | sort_by(-.total)' .codacy/tmp/codacy-cloud-overview.json

# Patterns with potential false positives (count > 0)
jq '[.overview.patterns[] | select(.potentialFalsePositives > 0) | {id, title, total, potentialFalsePositives, falsePositiveRatio: ((.potentialFalsePositives / .total) * 100 | round)}] | sort_by(-.falsePositiveRatio)' .codacy/tmp/codacy-cloud-overview.json

Identify noisy patterns using these criteria:

1. Wrong-language patterns — cross-reference pattern IDs against the discovered stack (Step 1 runs next, but the Cloud tools list from Step 0a gives a rough language picture). Pattern IDs often contain the target language (e.g., python., java., ruby.). Mark patterns for languages clearly not in the project.

2. Convention/style noise — patterns with very high issue counts (top 5% by count) in categories like CodeStyle, Documentation, or Comprehensibility are likely convention mismatches. Mark these as candidates.

3. High false-positive ratio patterns — patterns where a significant proportion of issues are flagged as potential false positives (e.g., potentialFalsePositives / total > 30%) are strong candidates for pre-disabling. These patterns are producing results that Codacy's own heuristics consider unreliable. Store the false positive data for use in Step 4b rule 7, where patterns with high false-positive ratios get additional scrutiny during local verification.

4. Never pre-disable Security patterns — security patterns are never marked for pre-disabling regardless of count or false-positive ratio. They are handled in Step 4 with full context.

5. Never pre-disable Critical/High severity patterns — these require local verification before any action.

Classify each noisy pattern into two lists:

• cloudNoiseLocal — noisy patterns from tools supported by the local Analysis CLI. These will be removed from .codacy/codacy.config.json immediately after init --auto in Step 2.
• cloudNoiseCloudOnly — noisy patterns from cloud-only tools (identified in Step 0b). These will be disabled via Cloud CLI during Step 6 (import), only if the import is performed.

Store both lists for use in Steps 2 and 6. This is a pre-filter — Step 4 will perform deeper noise evaluation on the local analysis results.

Step 1: Discover repository stack

Run discovery to understand the repository's languages, frameworks, and libraries:

codacy-analysis discover --output-format json --output .codacy/tmp/codacy-discover.json

Parse the output to understand:

• Languages present in the project
• Frameworks and libraries in use (e.g., React, Django, Spring Boot)
• This informs noise evaluation in Step 4 (e.g., knowing a project uses React means JSX-related patterns are relevant)

Note: The Codacy Cloud check already happened in Step 0.

Step 2: Initialize auto-tuned configuration

First, remove the existing .codacy/codacy.config.json if any (the backup copy was already stored in Step 0):

rm -f .codacy/codacy.config.json

Initialize with the broadest useful pattern set, filtered by the discovered stack:

codacy-analysis init --auto "Critical,High,Warning,Minor,AllSecurity,ErrorProne,Performance,BestPractice,UnusedCode,Compatibility,Complexity,Comprehensibility,CodeStyle,Documentation"

This filter means:

• Critical — Codacy-recommended (default) Critical-severity patterns
• AllSecurity — ALL Security-category patterns (including non-defaults)
• Everything else — Codacy-recommended (default) patterns at High, Warning, and Minor severity across all categories

The intent is to start broad and cut in Step 4 based on actual analysis data.

Apply Cloud noise pre-filter (Track A only): If cloudNoiseLocal patterns were identified in Step 0c, remove them from the newly created config now. For each pattern in cloudNoiseLocal, find it in .codacy/codacy.config.json under its tool's patterns array and remove it. This avoids wasting local analysis time on patterns already known to be noisy from Cloud production data. Track these removals for the summary (they are pre-filter changes, reported in patternChanges with reason referencing Cloud data).

Merge preserved configuration from the baseline stored in Step 0:

• Track A (Cloud): Merge from .codacy/tmp/codacy-remote-config.json
• Track B (local config): Merge from .codacy/tmp/codacy-previous-config.json
• Track C (no prior config): No merge needed

For the applicable track:

1. Read the newly created .codacy/codacy.config.json
2. Merge the global exclude array from the stored config (merge, don't replace — the new config may have its own excludes)
3. For each tool that exists in both stored and new config, merge its per-tool exclude array
4. For tools where useLocalConfigurationFile was true in the stored config and the tool still exists, restore that setting along with the localConfigurationFile path

Record broad-config metrics (internal — not the BEFORE reference for the summary):

# Count enabled patterns across all tools in the broad config
jq '[.tools[].patterns | length] | add' .codacy/codacy.config.json

# Count enabled tools in the broad config
jq '.tools | length' .codacy/codacy.config.json

Step 3: Run broad-config baseline analysis

Run analysis with the broad config to see the full issue landscape. This is an internal step for tuning decisions — the BEFORE reference for the user-facing summary comes from Step 0.

codacy-analysis analyze --install-dependencies --output-format json --output .codacy/tmp/codacy-baseline.json

Always use --output <file> to avoid broken JSON from stdout buffering.

Record broad-config metrics (used for tuning decisions in Step 4, not for the summary):

# Total issues in broad config
jq '.issues | length' .codacy/tmp/codacy-baseline.json

# Total runtime in milliseconds
jq '.metadata.durationMs' .codacy/tmp/codacy-baseline.json

Parse the issue distribution — this is the basis for all tuning decisions:

# Issues grouped by pattern (noise detection)
jq '[.issues | group_by(.patternId) | .[] | {patternId: .[0].patternId, toolId: .[0].toolId, severity: .[0].severity, category: .[0].category, count: length}] | sort_by(-.count)' .codacy/tmp/codacy-baseline.json

# Issues by severity
jq '.issues | group_by(.severity) | map({severity: .[0].severity, count: length})' .codacy/tmp/codacy-baseline.json

# Issues by category
jq '.issues | group_by(.category) | map({category: .[0].category, count: length})' .codacy/tmp/codacy-baseline.json

# Top 20 files by issue count
jq '[.issues | group_by(.filePath) | .[] | {filePath: .[0].filePath, count: length}] | sort_by(-.count) | .[0:20]' .codacy/tmp/codacy-baseline.json

# Per-tool results
jq '.toolResults | map({toolId, status, issueCount, durationMs, filesAnalyzed})' .codacy/tmp/codacy-baseline.json

Step 4: Smart noise evaluation and tuning

This is the core of the skill. Work through the baseline results using a structured, context-aware decision framework.

Cloud overview as a cross-reference (Track A only): If .codacy/tmp/codacy-cloud-overview.json exists (fetched in Step 0b), use its per-pattern issue counts and false positive data as a cross-reference against the local baseline. The cloud overview reflects the full repository analysis on Codacy (production truth), while the local baseline only covers files the local CLI can analyze. When a pattern shows high counts in the cloud overview but low counts locally, it may indicate cloud-only tool coverage — factor this into deduplication and disable decisions (don't disable a pattern that is still needed on Cloud). The false positive counts from the overview are especially valuable for rule 7 below — patterns with high false-positive ratios on Cloud are likely producing the same unreliable results locally.

4a. Establish the noise floor

Calculate the percentage of Critical+High issues (severity "Error" or "High") relative to total issues. This determines how aggressively to cut lower-priority categories:

The same logic applies proportionally to other low-priority categories (Comprehensibility, Compatibility) when critical issues dominate. Use judgment.

4b. Pattern-level decisions

For each pattern in the baseline results, sorted by issue count (highest first), apply this priority chain:

1. Security patterns must cover every security concern. Any pattern with category == "Security" stays enabled UNLESS another active pattern from a different tool already covers the same semantic concern (see rule 2). If a security pattern is noisy and no deduplication applies, exclude specific files instead of disabling the pattern.

2. Cross-tool deduplication of overlapping patterns. When multiple tools flag the same semantic concern (e.g., "hardcoded secrets", "SQL injection", "XSS", "unused imports"), identify the overlap by comparing issue messages, affected files, and affected lines. For the overlapping pair:

   • Keep the pattern from the more specialized or precise tool (e.g., Semgrep security rules over Biome generic checks; a dedicated SAST scanner over a general linter)
   • If precision is comparable, keep the pattern from the tool that is more actively used in the project (more patterns enabled, used in CI, has a local config file)
   • Disable the redundant pattern from the other tool
   • This is NOT removing a concern — it is deduplicating. The concern remains covered by the kept pattern.
   • This applies to ALL categories, not just Security — e.g., two tools both flagging "unused imports"
   • Document in the change log: which pattern was kept, which was disabled, and why the kept pattern is the better source

3. NEVER disable valid Critical/High issues. Patterns with severity == "Error" or severity == "High" that are finding real problems must stay. If they appear to be false positives, exclude the offending files rather than disabling the pattern.

4. Wrong stack → disable. Cross-reference with the discover output from Step 1. Patterns for languages or frameworks not present in the project are pure noise. Examples:

   • Python security patterns in a JavaScript-only project
   • Apex patterns from PMD7 in a Java-only project
   • Semgrep rules for languages not in the repo

   Remove these patterns from the config.

5. Noise floor → disable. Apply the decisions from 4a. If the noise floor says CodeStyle/Documentation at Minor severity should go, remove those patterns.

6. Convention mismatch → disable. If a pattern flags something that >80% of the codebase does consistently, the pattern contradicts the project's established conventions. Examples:

   • Tabs-vs-spaces rules when the entire project uses the "wrong" style consistently
   • Naming convention rules that don't match the project's established naming

7. False-positive prone → disable. Use the Cloud overview's false positive data (from Step 0c) as the primary signal: patterns where potentialFalsePositives / total > 30% are strong disable candidates — Codacy's own heuristics are flagging a significant share of their results as unreliable. For these patterns, review a sample of actual instances in the code to confirm. If the local instances confirm the pattern is producing low-value results for this codebase, disable it. For patterns without Cloud false-positive data, check if the pattern is known to produce high false-positive rates in general. Only disable after verifying the hits are genuinely not useful for this codebase.

8. Parameter tuning over disabling. Before disabling a valuable pattern, check if it has configurable parameters:

   • Lizard complexity thresholds — raise to match the codebase's actual complexity profile
   • Line length limits — set to the project's observed maximum
   • Other threshold-based rules — adjust to reduce false hits while keeping the rule active

   Tuning preserves coverage while reducing noise.

9. File exclusion over disabling. When a pattern is valid but fires on files where it doesn't apply, exclude the files rather than disabling the pattern. The pattern stays active for real source code.

4c. File-level evaluation

Review the top files by issue count from the baseline results (Step 3). Exclusions must be strictly data-driven — only add exclusions for files/paths that actually appear in the baseline results and are producing meaningful noise.

Important: The analysis CLI already respects .gitignore. Files matched by .gitignore are never analyzed, so they do NOT need exclusion in the Codacy config. Do NOT preemptively add exclusions for paths like dist/**, node_modules/**, build/**, or coverage/** unless they are actually appearing in the results with issues.

Process for each noisy file/path in the top-N by issue count:

1. Check if the file represents generated code (e.g., *.generated.ts, routeTree.gen.ts, auto-generated API clients, .pb.go files)
2. Check if the file is vendored or third-party code committed to the repo (e.g., .yarn/releases/, vendor/ that is not gitignored)
3. Check if the file is build output that was committed (not gitignored)
4. Check if the file is a test fixture, snapshot, or mock data that produces false positives from a specific tool
5. If any of the above apply, add to the appropriate exclusion:
   • Generated/vendored/build output that affect all tools: add to the global exclude array
   • Files noisy for a specific tool only: add to that tool's per-tool exclude array (e.g., markdownlint hitting CHANGELOG.md, Stylelint hitting committed vendor CSS)
6. If the file is legitimate source code, do NOT exclude it — address the noise through pattern-level tuning (rules 1-9 above) instead

Every exclusion must be justified by actual results. Do NOT maintain a prescriptive checklist of "always exclude these paths."

4d. Tool-level evaluation

For each tool in the config:

• Irrelevant to the stack? If the discover output shows no files for a tool's target language, remove the entire tool entry. Examples: RuboCop in a project with no Ruby, Checkstyle in a project with no Java.
• Failed to run? Check toolResults[].status. If "failed", check errors[] for the reason. Keep the tool in the config (it may work after fixing the root cause) but note it in the summary.
• Zero issues? A tool with zero issues is not noise — it's either confirming code quality or not finding its target files. Leave it enabled unless it's clearly irrelevant to the stack.

4e. Lost patterns recovery

Compare the baseline analysis results (from Step 0) against the current .codacy/codacy.config.json after tuning. For each pattern that:

• Found issues in the baseline analysis (.codacy/tmp/codacy-remote-results.json for Track A, .codacy/tmp/codacy-previous-results.json for Track B)
• Has category == "Security" OR severity == "Error" OR severity == "High"
• Is NOT present in the current config (was excluded by init --auto or removed during tuning)

→ Add it back to the config under the appropriate tool. This is an internal preservation step — restored patterns are NOT reported as changes in the summary (they were already in the baseline config, so from the user's perspective nothing changed).

Skip this step for Track C (no prior config — nothing to recover).

Apply all changes

Edit .codacy/codacy.config.json with all decisions from 4a–4e. Track every change for the summary in Step 5:

• For each disabled pattern: record the patternId, action "disabled", and reason
• For each tuned pattern: record the patternId, action "updated", old/new parameters, and reason
• For each removed tool: record it in toolChanges
• For each added tool: record it in toolChanges
• For each new file exclusion added: record it in fileExclusions
• Restored patterns (Step 4e) are NOT included in the summary — they preserve the status quo

Security guardrail — mandatory check before saving. Before writing the updated config, verify that every security concern (e.g., hardcoded secrets, SQL injection, XSS, path traversal, open redirects) is still covered by at least one active pattern with category == "Security". A security pattern may be disabled only if another active pattern from a different tool covers the same concern (cross-tool deduplication per rule 2 in Step 4b). If any concern has lost all coverage, re-enable the most precise pattern for that concern.

4f. Validation pass

Run analysis with the tuned config to validate the improvement:

codacy-analysis analyze --install-dependencies --output-format json --output .codacy/tmp/codacy-tuned.json

Record AFTER metrics:

# Enabled patterns after tuning
jq '[.tools[].patterns | length] | add' .codacy/codacy.config.json

# Enabled tools after tuning
jq '.tools | length' .codacy/codacy.config.json

# Total issues after tuning
jq '.issues | length' .codacy/tmp/codacy-tuned.json

# Runtime after tuning
jq '.metadata.durationMs' .codacy/tmp/codacy-tuned.json

Validate:

• Issues should have decreased meaningfully vs the broad-config baseline (Step 3).
• If issues increased or didn't decrease meaningfully (<20% reduction), review the tuning decisions and re-iterate once: go back to "Do AI analysis of the results" (Step 4b) with the updated results, apply further changes, and re-run this validation.
• The goal is that every remaining issue is worth looking at.

Step 5: Show local results

5a. Generate summary JSON

Write .codacy/configure-codacy-summary.json with the before/after metrics, a detailed change log, and supporting context. Runtime values are in milliseconds (raw durationMs from the CLI).

The before values come from Step 0 (the baseline captured before any changes — Cloud config for Track A, local config for Track B, or unconfigured for Track C). The after values come from Step 4f (the final tuned configuration). If no initial config existed (Track C), before values reflect an unconfigured state — use null for metrics that cannot be measured (e.g., runtime) and 0 for counts (patterns, tools, issues).

{
  "summary": {
    "enabledPatterns": { "before": 1000, "after": 300 },
    "enabledTools": { "before": 34, "after": 15 },
    "issues": { "before": 7000, "after": 550 },
    "analysisRuntime": { "before": 240000, "after": 60000 }
  },
  "toolChanges": [
    {
      "toolId": "Biome",
      "action": "disabled",
      "reason": "Project uses ESLint9 with local config; Biome is redundant and produced 16K false positives in TypeScript",
      "patternsAffected": 232
    },
    {
      "toolId": "markdownlint",
      "action": "enabled",
      "reason": "New tool added for Markdown linting; README.md excluded per-tool to avoid inline HTML noise",
      "patternsAffected": 43
    }
  ],
  "patternChanges": [
    {
      "patternId": "Semgrep_python.lang.security.audit.xss.template-injection",
      "toolId": "Semgrep",
      "action": "disabled",
      "reason": "Wrong stack — pattern for Python; project is JavaScript-only",
      "delta": -45,
      "parameters": []
    },
    {
      "patternId": "Lizard_ccn-medium",
      "toolId": "Lizard",
      "action": "updated",
      "reason": "Raised threshold from 10 to 20 to match codebase complexity profile",
      "delta": -120,
      "parameters": [
        { "id": "threshold", "before": "10", "after": "20" }
      ]
    },
    {
      "patternId": "Semgrep_codacy.javascript.security.hard-coded-password",
      "toolId": "Semgrep",
      "action": "enabled",
      "reason": "New High Security pattern — detects hardcoded passwords; found 101 issues across 41 files for review",
      "delta": 101,
      "parameters": []
    }
  ],
  "fileExclusions": {
    "global": ["src/routeTree.gen.ts"],
    "perTool": {
      "markdownlint": ["README.md"]
    }
  },
  "securityCoverage": {
    "deduplication": [
      "Semgrep slack-webhook-url disabled — same concern covered by Semgrep detected-slack-webhook"
    ],
    "newCoverage": [
      "unsafe-dynamic-method (Critical, 5 issues)",
      "open-redirect-from-function (Critical, 4 issues)",
      "hard-coded-password (High, 101 issues)"
    ],
    "noisyButKept": [
      "Semgrep hard-coded-password (101 issues) — kept per security guardrail, file exclusions preferred over disabling"
    ]
  },
  "keyImprovements": [
    "Lizard complexity thresholds tuned to match mature React SPA — 576 fewer noise issues while keeping genuinely complex functions flagged",
    "6 new security patterns detecting open redirects, unsafe dynamic methods, hardcoded passwords, and React href injection (119 new findings)",
    "markdownlint added for Markdown quality",
    "Checkov expanded from 6 to 1358 IaC security patterns"
  ],
  "localConfigTools": [
    {
      "toolId": "ESLint9",
      "configFile": "eslint.config.js",
      "issueCount": 42,
      "note": "Issues from this tool are controlled by the project's ESLint config, not Codacy patterns. Edit eslint.config.js to reduce noise."
    }
  ],
  "codacyYaml": null,
  "importResults": null,
  "cloudVerification": null
}

Field reference:

summary — before/after metrics (same as before).

toolChanges — one entry per tool added or removed. Each entry:

• toolId — the tool identifier
• action — "enabled" (tool was added) or "disabled" (tool was removed)
• reason — why the tool was added or removed
• patternsAffected — number of patterns in the tool that was added/removed

patternChanges — one entry per individual pattern change. Each entry:

• patternId — the pattern identifier
• toolId — which tool this pattern belongs to
• action — "enabled", "disabled", or "updated"
• reason — why this change was made (auditability)
• delta — the change in issue count for this pattern, comparing INITIAL state (Step 0) vs FINAL tuned state (Step 4f). If the pattern was not enabled in the initial config, the delta is the full issue count from the tuned analysis (positive if newly enabled, 0 if disabled during tuning).
• parameters — parameter changes (empty array if not applicable)

Do NOT include individual pattern entries inside patternChanges for patterns that were added/removed as part of a whole tool change. Those are covered by the toolChanges entry. patternChanges is for surgical, individual pattern changes within a tool that remains enabled.

Do NOT include "restored" patterns here. Restoration (Step 4e) is an internal mechanism to ensure patterns from the baseline config that had Critical/High/Security results aren't lost by init --auto. It preserves what was already there — it's not a user-facing change.

fileExclusions — only lists new exclusions added during this tuning run. Exclusions already present in the baseline config are preserved automatically but not listed here (they are not changes).

• global — new global exclusion globs added during tuning
• perTool.<toolId> — new per-tool exclusion globs added during tuning

Omit fileExclusions entirely if no new exclusions were added.

securityCoverage — documents how security concerns are handled:

• deduplication — security patterns that were disabled because another pattern covers the same concern (list the kept pattern and the disabled one)
• newCoverage — new security patterns enabled that were NOT in the baseline config, with issue count
• noisyButKept — security patterns that are noisy but were kept active per the security guardrail

keyImprovements — array of 3–6 human-readable sentences summarizing the most impactful improvements. Focus on what changed and the quantitative impact. These should be suitable for presenting to the user as a summary.

localConfigTools — array of tools that have useLocalConfigurationFile: true. Each entry includes:

• toolId — the tool identifier
• configFile — path to the project's config file used by this tool
• issueCount — number of issues this tool produced in the tuned analysis
• note — human-readable explanation that noise from this tool is governed by the project's own config, not Codacy patterns

Empty array if no tools use local configuration files.

codacyYaml — string containing the generated .codacy.yaml file content if file exclusions were added (see Step 5c). null if no .codacy.yaml was generated.

importResults — null until Step 6 runs. If Cloud import is performed, this is populated with the results (see Step 6f for the schema).

cloudVerification — null unless Step 6e runs. Contains post-import Cloud verification results: whether reanalysis completed, issue counts before/after, patterns disabled on Cloud, and warnings. See Step 6e for the schema.

5b. Present local results

Display a clear before/after summary to the user:

1. Metrics table — enabled patterns, enabled tools, total issues, and runtime (before vs after, with reduction %). The "before" is the baseline from Step 0. If no prior config existed (Track C), note "No prior configuration" in the before column.
2. Tool changes — tools added or removed, with reasons
3. Top pattern changes by impact — the patternChanges entries with the largest |delta|, sorted descending
4. New security coverage — new security patterns that were not in the baseline config
5. Security deduplication — any security patterns that were deduplicated (which was kept, which was disabled)
6. Key improvements — the keyImprovements array from the summary JSON, presented as a bulleted list
7. Warnings — failed tools, Semgrep parsing errors, tools with 0 files matched, any other issues encountered
8. Local config file limitations — for tools with useLocalConfigurationFile: true, note that their issues are controlled by the project's own config file (e.g., .eslintrc, .rubocop.yml), not by Codacy's managed patterns. Noise from these tools can only be reduced by editing the project's own config. List each such tool and the number of issues it produced, so the user knows where to look if they want further noise reduction.

5c. Generate .codacy.yaml for Cloud file exclusions

File exclusions in .codacy/codacy.config.json only apply to local analysis. Codacy Cloud does not support importing file exclusions via the API or CLI. To apply file exclusions on Codacy Cloud, they must be defined in a .codacy.yaml file committed to the repository root (see Codacy configuration file).

If any new file exclusions were added during tuning (global or per-tool), generate a .codacy.yaml file:

1. Read the existing .codacy.yaml from the repo root (if it exists) to preserve any existing configuration
2. Merge the new exclusions with any existing exclude_paths and per-engine exclude_paths
3. Write the updated .codacy.yaml to the repo root

The .codacy.yaml format uses Java glob syntax:

---
exclude_paths:
  - "src/routeTree.gen.ts"
  - "vendor/**"
engines:
  markdownlint:
    exclude_paths:
      - "CHANGELOG.md"
  stylelint:
    exclude_paths:
      - "assets/vendor/**"

Engine name mapping: The engine names in .codacy.yaml may differ from the tool IDs in .codacy/codacy.config.json. Use the tool ID in lowercase as the engine name (e.g., ESLint9 → eslint9, Semgrep → semgrep, markdownlint → markdownlint). If unsure, the tool ID in lowercase is the safe default.

4. Store the full content of the generated .codacy.yaml as a string in the codacyYaml field of the summary JSON
5. Present to the user:
   • Note that file exclusions cannot be imported to Codacy Cloud via API
   • The .codacy.yaml file has been created/updated in the repo root
   • The user should commit and push this file to apply the exclusions on Codacy Cloud
   • Codacy Cloud always reads .codacy.yaml from the default branch

If no new file exclusions were added, skip this step and leave codacyYaml as null.

Step 6: Cloud import (conditional)

6a. Check Cloud status

If the repo is NOT on Codacy Cloud (from Step 0): skip this step entirely. Note in the results: "The configuration is ready for local analysis. To use it on Codacy Cloud, first add the repository to Codacy."

6b. Determine import behavior

Based on invocation mode (see "Invocation modes" section):

1. Auto-import mode (user invoked with "import" argument): Proceed directly to 6c.
2. Interactive mode (default): Use AskUserQuestion to ask the user: "Want to update these configurations in Codacy Cloud?" If the user declines, skip to Step 7.
3. Local-only mode: Already handled in 6a (skipped).

6c. Cloud-only tools noise evaluation

Only if cloud-only tools had issues fetched in Step 0 (.codacy/tmp/codacy-remote-cloud-results.json exists):

Apply the same noise evaluation framework from Step 4 to the cloud-fetched issues. For each noisy pattern from a cloud-only tool:

• Check if its parameters can be tweaked t

…(truncated)