Imported from jfrog/ai-agent-examples (onboarding-workflows/skills/detect-existing-patterns/SKILL.md). Install upstream with npx skills add jfrog/ai-agent-examples --skill detect-existing-patterns. Copyright stays with the author.

Detect Existing Naming Patterns

Queries the JFrog Platform for existing projects and repositories, analyzes their naming conventions, and presents the user with a choice between the detected pattern and the standard onboarding pattern.

API transport

Prefer jf api per ../../../platform-features/skills/jfrog-cli/jf-api-patterns.md after jf config and platform confirmation. Examples below use curl as fallback.

Inputs

• JFROG_URL -- Platform base URL (for curl fallback; derive from jf config show when using CLI)
• JFROG_ACCESS_TOKEN -- Admin token (for curl fallback only)
• new_project_key -- the project key about to be onboarded (used to generate examples)
• new_ecosystems -- list of ecosystems for the new project (used to generate examples)
• state_project_key -- (optional) the state/system project key to exclude from analysis (default: system)

Output

A naming_pattern object with these fields (passed to jfrog-create-repos):

The standard (default) pattern produces names like {key}-npm-local, {key}-npm-remote, {key}-npm.

When to Skip

• If the platform has no existing projects (or only the state project), skip detection entirely and use the standard pattern. Do not ask the user anything.
• If no repos can be found across existing projects, skip detection and use the standard pattern.

Workflow

0. Load credentials

if [ -z "$JFROG_URL" ] || [ -z "$JFROG_ACCESS_TOKEN" ]; then
  if [ -f .env ]; then
    set -a; source .env; set +a
  fi
fi

1. Fetch all existing projects

Run with required_permissions: ["full_network"].

BODY=/tmp/all-projects.json
CODE=/tmp/all-projects.code
jf api /access/api/v1/projects >"$BODY" 2>"$CODE"
HTTP_CODE=$(tr -d '\r\n' < "$CODE")

if [ "$HTTP_CODE" = "200" ]; then
  PROJECT_KEYS=$(jq -r '.[].project_key' "$BODY")
else
  echo "WARN: Could not list projects (HTTP $HTTP_CODE). Skipping pattern detection."
  # Use standard pattern
fi

Fallback: curl -s -o /tmp/all-projects.json -w "%{http_code}" -H "Authorization: Bearer $JFROG_ACCESS_TOKEN" "$JFROG_URL/access/api/v1/projects"

Filter out the state project (default key system, or the value of state_project_key input):

PROJECT_KEYS=$(echo "$PROJECT_KEYS" | grep -v "^${STATE_PROJECT_KEY}$")

If PROJECT_KEYS is empty after filtering, skip detection -- use the standard pattern and exit this skill.

2. Fetch repositories per project

Run with required_permissions: ["full_network"]. Collect all repo keys grouped by project. Add sleep 1 between iterations to avoid rate limiting.

ALL_REPOS=""
for PROJ in $PROJECT_KEYS; do
  RB=/tmp/repos-${PROJ}.json
  RC=/tmp/repos-${PROJ}.code
  jf api "/artifactory/api/repositories?project=${PROJ}" >"$RB" 2>"$RC"
  HTTP_CODE=$(tr -d '\r\n' < "$RC")

  if [ "$HTTP_CODE" = "200" ]; then
    REPO_KEYS=$(jq -r '.[].key' "$RB")
    for RK in $REPO_KEYS; do
      ALL_REPOS="${ALL_REPOS}${PROJ}|${RK}\n"
    done
  fi
  sleep 1
done

If no repos were collected, skip detection -- use the standard pattern and exit this skill.

3. Analyze naming patterns

For each project's repos, decompose repo names to detect the naming convention. The analysis checks three characteristics: separator, component order, and virtual suffix.

Known tokens for matching

• Ecosystem tokens: npm, maven, pypi, go, docker, helm, nuget, generic, gradle, cargo, conan, conda, gems, gitlfs, bower, cocoapods, pub, python, pip
• Type tokens: local, remote, virtual
• Type abbreviations: loc -> local, rem -> remote, virt -> virtual

Detection algorithm

For each PROJECT_KEY | REPO_KEY pair:

1. Strip the project key from the repo name (it may appear as a prefix or elsewhere). Record the position where it appears.

2. Identify the separator. Try each candidate separator (-, _, .) and split the repo name. The separator that produces segments matching the most known tokens (project key, ecosystem, type) wins. If all three produce equal matches, default to -.

3. Split the repo name using the winning separator.

4. Classify each segment as one of:

   • project_key -- matches the project key
   • ecosystem -- matches a known ecosystem token
   • type -- matches a known type token or abbreviation
   • unknown -- does not match any known token

5. Record the component order -- the sequence of classified segments. For example: ["project_key", "ecosystem", "type"] or ["ecosystem", "project_key", "type"].

6. Record the type mapping -- what string was used for each logical type. For example, if loc was used instead of local, record {"local": "loc"}.

7. Record the ecosystem mapping -- what string was used. For example, if python was used instead of pypi, record {"pip": "python"}.

8. Determine the virtual suffix. Look at repos with rclass: virtual (from the repo list API response which includes type field). Compare their names to corresponding local/remote repos:

   • If the virtual repo name has no type segment (e.g., myproj-npm): virtual_suffix = "" (empty)
   • If it has an explicit type segment (e.g., myproj-npm-virtual): virtual_suffix = that segment

Aggregation across repos

After analyzing all repos:

1. Count separator usage -- pick the separator with the highest frequency.
2. Count component order -- pick the order with the highest frequency.
3. Count virtual suffix -- pick the suffix with the highest frequency.
4. Build the type_map and eco_map from the most common mappings.
5. Compute confidence:
   • High: 80%+ of repos match the same pattern
   • Medium: 50-79% match the most common pattern
   • Low: less than 50% match any single pattern

4. Present options to the user

Use the AskQuestion tool to present the choice. Before asking, display context about what was found.

Generate example names

Using the user's new_project_key and new_ecosystems, generate concrete example repo names for both options.

Detected pattern example -- apply the detected separator, order, virtual_suffix, type_map, and eco_map:

# If detected pattern is: {project_key}_{ecosystem}_{type}, virtual_suffix="virtual"
# For project "webapp", ecosystem "npm":
#   webapp_npm_local
#   webapp_npm_remote
#   webapp_npm_virtual

Standard pattern example -- always {key}-{ecosystem}-{type}, no virtual suffix, pip->pypi:

# For project "webapp", ecosystem "npm":
#   webapp-npm-local
#   webapp-npm-remote
#   webapp-npm

Present the choice

Display a summary message to the user:

Found {N} existing projects with {M} repositories on the platform.

Detected naming pattern (confidence: {high/medium/low}):
  Separator: "{sep}"
  Order: {project_key}{sep}{ecosystem}{sep}{type}
  Virtual repos: {description of virtual suffix}
  {Any notable ecosystem mappings, e.g. "pip ecosystem uses 'python' in repo names"}

Existing examples:
  {list 3-5 actual existing repo names as examples}

Examples for your new project "{new_project_key}" with ecosystems [{new_ecosystems}]:

  Option A -- Follow existing pattern:
    {generated detected-pattern examples, one line per repo}

  Option B -- Standard pattern:
    {generated standard-pattern examples, one line per repo}

Then use AskQuestion:

AskQuestion:
  id: naming_pattern_choice
  prompt: "Which naming pattern would you like to use for the new repositories?"
  options:
    - id: detected
      label: "Option A -- Follow existing pattern ({detected_pattern_summary})"
    - id: standard
      label: "Option B -- Standard pattern ({key}-{ecosystem}-{type})"

Detected pattern matches standard

If the detected pattern is identical to the standard pattern (separator = -, order = [project_key, ecosystem, type], virtual_suffix = "", eco_map = {"pip": "pypi"}), do not ask the user. Simply report:

Existing repositories already follow the standard naming pattern. Proceeding with standard naming.

And return the standard naming_pattern.

5. Return the naming_pattern

Based on the user's choice (or the default when detection is skipped or the patterns match), return the naming_pattern object:

Standard pattern (default):

separator: "-"
virtual_suffix: ""
type_map:
  local: "local"
  remote: "remote"
eco_map:
  pip: "pypi"

Detected pattern (example):

separator: "_"
virtual_suffix: "virtual"
type_map:
  local: "local"
  remote: "remote"
eco_map:
  pip: "python"

The orchestration skill stores this and passes it to jfrog-create-repos for each project.

Repo Name Generation

Given a naming_pattern, generate a repo key as follows:

SEP="${NAMING_SEPARATOR:--}"

# Map ecosystem name to repo token
ECO_TOKEN="$ECOSYSTEM"
# Apply eco_map overrides (e.g., pip -> pypi)
case "$ECOSYSTEM" in
  pip) ECO_TOKEN="${ECO_MAP_PIP:-pypi}" ;;
esac

# Local repo
LOCAL_TYPE="${TYPE_MAP_LOCAL:-local}"
REPO_KEY="${PROJECT_KEY}${SEP}${ECO_TOKEN}${SEP}${LOCAL_TYPE}"

# Remote repo
REMOTE_TYPE="${TYPE_MAP_REMOTE:-remote}"
REPO_KEY="${PROJECT_KEY}${SEP}${ECO_TOKEN}${SEP}${REMOTE_TYPE}"

# Virtual repo
VIRTUAL_SUFFIX="${NAMING_VIRTUAL_SUFFIX:-}"
if [ -z "$VIRTUAL_SUFFIX" ]; then
  REPO_KEY="${PROJECT_KEY}${SEP}${ECO_TOKEN}"
else
  REPO_KEY="${PROJECT_KEY}${SEP}${ECO_TOKEN}${SEP}${VIRTUAL_SUFFIX}"
fi

Error Handling

• If the projects API fails, log a warning and fall back to the standard pattern. Do not abort onboarding.
• If the repos API fails for a specific project, skip that project's repos in the analysis and continue.
• If the detected pattern has low confidence, include a note in the presentation: "The detected pattern has low confidence because existing repos use mixed naming conventions."
• Pattern detection is best-effort; it must never block or abort the onboarding workflow.