Merge pull request #6412 from ggiguash/analyze-ci-tmp-dir

USHIFT-6760: Update analyze-ci work directory to have a date suffix

Author: openshift-merge-bot[bot]

.claude/commands/analyze-ci-create-bugs.md

@@ -15,7 +15,7 @@ allowed-tools: Bash, Read, Write, Glob, Grep, Agent, mcp__jira__jira_search, mcp
 ## Description
 Reads individual job analysis reports produced by `analyze-ci-for-release` or `analyze-ci-for-pull-requests` and creates JIRA bugs in USHIFT for legitimate test failures. Operates in **dry-run mode by default** - it shows what bugs would be created without actually creating them. Use `--create` to perform actual issue creation.
 
-This command does NOT re-analyze CI jobs. It consumes existing job analysis files from `/tmp/analyze-ci-claude-workdir/`.
+This command does NOT re-analyze CI jobs. It consumes existing job analysis files from `${WORKDIR}/`.
 
 ## Arguments
 - `$ARGUMENTS` (required): Source identifier, optionally followed by `--create`
@@ -27,7 +27,7 @@ This command does NOT re-analyze CI jobs. It consumes existing job analysis file
 
 ## Prerequisites
 
-- Job analysis files must already exist in `/tmp/analyze-ci-claude-workdir/`:
+- Job analysis files must already exist in `${WORKDIR}/`:
   - For releases: `analyze-ci-release-<release>-job-*.txt` (produced by `/analyze-ci-for-release`)
   - For PRs: `analyze-ci-prs-job-*-pr<number>-*.txt` (produced by `/analyze-ci-for-pull-requests`)
 - Each job file must contain a `--- STRUCTURED SUMMARY ---` block (see below)
@@ -54,28 +54,35 @@ FINISHED: <job finish date in YYYY-MM-DD format>
 
 If a job file lacks this block, it is skipped with a warning.
 
+## Work Directory
+
+Set once at the start and reference throughout:
+```bash
+WORKDIR=/tmp/analyze-ci-claude-workdir.$(date +%y%m%d)
+```
+
 ## Implementation Steps
 
 ### Step 1: Parse Arguments and Locate Job Files
 
 **Actions**:
 1. Parse `$ARGUMENTS` to extract `<source>` and detect `--create` flag
 2. Determine mode: if `--create` is present, set `MODE=create`; otherwise `MODE=dry-run`
-3. Run `mkdir -p /tmp/analyze-ci-claude-workdir` using the `Bash` tool
+3. Run `WORKDIR=/tmp/analyze-ci-claude-workdir.$(date +%y%m%d) && mkdir -p ${WORKDIR}` using the `Bash` tool
 4. Determine source type and locate job files:
 
    **a) Release version** (e.g., `4.22`, `main`, `4.19`):
-   - Glob for: `/tmp/analyze-ci-claude-workdir/analyze-ci-release-<release>-job-*.txt`
+   - Glob for: `${WORKDIR}/analyze-ci-release-<release>-job-*.txt`
    - Set `SOURCE_TYPE=release`, `SOURCE_LABEL="release <release>"`
 
    **b) PR number** (e.g., `pr-6396`, `pr6396`):
    - Extract the numeric PR number (strip `pr-` or `pr` prefix)
-   - Glob for: `/tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-*-pr<number>-*.txt`
+   - Glob for: `${WORKDIR}/analyze-ci-prs-job-*-pr<number>-*.txt`
    - Set `SOURCE_TYPE=pr`, `SOURCE_LABEL="PR #<number>"`
 
    **c) Rebase PR shorthand** (e.g., `rebase-release-4.22`):
    - Extract the release version from the shorthand (e.g., `4.22`)
-   - Glob for all PR job files: `/tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-*.txt`
+   - Glob for all PR job files: `${WORKDIR}/analyze-ci-prs-job-*.txt`
    - Read each file's STRUCTURED SUMMARY and find files where JOB_NAME contains the release version (e.g., `release-4.22` or `main`) OR where the JOB_URL contains the release branch
    - Alternatively, check the PR summary file (`analyze-ci-prs-summary.*.txt`) to find the PR number for the given rebase release
    - Set `SOURCE_TYPE=pr`, `SOURCE_LABEL="rebase PR for <release> (PR #<number>)"`
@@ -152,24 +159,26 @@ If a job file lacks this block, it is skipped with a warning.
 
 For each bug candidate, run ALL of the following searches. Each search is MANDATORY — do not skip any.
 
-**Search A — Keyword search**:
+**Search A — Keyword search (multiple focused queries)**:
 1. Extract 2-4 distinctive keywords from the error signature (avoid generic words like "error", "failed", "test")
-2. Run:
+2. Run **2-3 separate searches in parallel**, each using 1-2 distinctive keywords. Do NOT put all keywords into a single `text ~` query — Jira requires all terms to match, so queries with 3+ keywords are fragile and miss issues that use slightly different wording. Instead, pick the most distinctive terms (tool names, hyphenated identifiers, error codes) and search for them independently.
    ```python
-   mcp__jira__jira_search(
-     jql='((project = OCPBUGS AND component = MicroShift) OR project = USHIFT) AND text ~ "<keywords>" AND status not in (Closed, Verified) ORDER BY created DESC',
-     limit=5
-   )
+   # Example for error signature: "TLS Scanner image pull failure (ErrImagePull registry.ci.openshift.org/ocp/4.22:tls-scanner-tool on arm64)"
+   # Search A1: most distinctive identifier
+   mcp__jira__jira_search(jql='... AND issuetype = Bug AND text ~ "tls-scanner" ...', limit=5)
+   # Search A2: different distinctive term
+   mcp__jira__jira_search(jql='... AND issuetype = Bug AND text ~ "tls-scanner-tool" ...', limit=5)
    ```
+3. Merge and deduplicate results from all A-series queries before proceeding
 
 **Search B — Test case ID search (MANDATORY when IDs are present)**:
 Extract ALL numeric IDs from the error signature that could be test case references (typically 4-6 digit numbers like `68256`). For EACH numeric ID found, run TWO separate searches:
 ```text
 # Search B1: bare number
-jql: ... AND text ~ "68256" AND status not in (Closed, Verified) ...
+jql: ... AND issuetype = Bug AND text ~ "68256" AND status not in (Closed, Verified) ...
 
 # Search B2: OCP-prefixed form (OpenShift Polarion convention)
-jql: ... AND text ~ "OCP-68256" AND status not in (Closed, Verified) ...
+jql: ... AND issuetype = Bug AND text ~ "OCP-68256" AND status not in (Closed, Verified) ...
 ```
 **Why both forms are required**: Jira's text indexer treats `OCP-68256` as a single token, so `text ~ "68256"` will NOT match issues containing `OCP-68256`, and vice versa. Skipping either form WILL cause missed duplicates.
 
@@ -181,7 +190,7 @@ jql: ... AND text ~ "OCP-68256" AND status not in (Closed, Verified) ...
 After completing searches A and B, run an additional keyword search against closed/verified issues to detect potential regressions:
 ```python
 mcp__jira__jira_search(
-  jql='((project = OCPBUGS AND component = MicroShift) OR project = USHIFT) AND text ~ "<keywords>" AND status in (Closed, Verified) ORDER BY updated DESC',
+  jql='((project = OCPBUGS AND component = MicroShift) OR project = USHIFT) AND issuetype = Bug AND text ~ "<keywords>" AND status in (Closed, Verified) ORDER BY updated DESC',
   limit=5
 )
 ```
@@ -381,7 +390,7 @@ For each candidate where user chose "reopen":
 ### Step 8: Generate Results Report
 
 **Actions**:
-1. Save report to `/tmp/analyze-ci-claude-workdir/analyze-ci-create-bugs-<source>.<timestamp>.txt`
+1. Save report to `${WORKDIR}/analyze-ci-create-bugs-<source>.<timestamp>.txt`
 2. Display summary to user:
 
 **Dry-run report format**:
@@ -418,7 +427,7 @@ CANDIDATES
 To create these bugs, run:
   /analyze-ci-create-bugs <source> --create
 
-Report saved: /tmp/analyze-ci-claude-workdir/analyze-ci-create-bugs-<source>.<timestamp>.txt
+Report saved: ${WORKDIR}/analyze-ci-create-bugs-<source>.<timestamp>.txt
 ═══════════════════════════════════════════════════════════════
 ```
 
@@ -456,7 +465,7 @@ SUMMARY
   Reopened: N
   Failed: N
 
-Report saved: /tmp/analyze-ci-claude-workdir/analyze-ci-create-bugs-<source>.<timestamp>.txt
+Report saved: ${WORKDIR}/analyze-ci-create-bugs-<source>.<timestamp>.txt
 ═══════════════════════════════════════════════════════════════
 ```
 
@@ -491,7 +500,7 @@ Resolves the rebase PR for release 4.22, then interactively creates bugs.
 /analyze-ci-create-bugs 4.19
 ```
 ```text
-Error: No job analysis files found at /tmp/analyze-ci-claude-workdir/analyze-ci-release-4.19-job-*.txt
+Error: No job analysis files found at ${WORKDIR}/analyze-ci-release-4.19-job-*.txt
 
 Run the analysis first:
   /analyze-ci-for-release 4.19
@@ -502,15 +511,15 @@ Run the analysis first:
 /analyze-ci-create-bugs pr-9999
 ```
 ```text
-Error: No job analysis files found at /tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-*-pr9999-*.txt
+Error: No job analysis files found at ${WORKDIR}/analyze-ci-prs-job-*-pr9999-*.txt
 
 Run the analysis first:
   /analyze-ci-for-pull-requests
 ```
 
 ## Notes
 
-- This command does NOT run CI analysis — it only consumes existing analysis files from `/tmp/analyze-ci-claude-workdir`
+- This command does NOT run CI analysis — it only consumes existing analysis files from `${WORKDIR}`
 - Supports two file naming patterns:
   - Release jobs: `analyze-ci-release-<release>-job-*.txt` (from `/analyze-ci-for-release`)
   - PR jobs: `analyze-ci-prs-job-*-pr<number>-*.txt` (from `/analyze-ci-for-pull-requests`)

.claude/commands/analyze-ci-for-prow-job.md

@@ -1,7 +1,7 @@
 ---
 name: Analyze CI for a Prow Job
 argument-hint: <prow-job-url>
-description: Analyze a MicroShift Prow CI Test Job execution
+description: Download Prow job artifacts, identify root cause of failure, and produce a structured error report
 allowed-tools: Skill, Bash, Read, Write, Glob, Grep, Agent
 ---
 
@@ -75,12 +75,19 @@ ${TMP}/artifacts/${TEST_NAME}/openshift-microshift-infra-sos-aws/artifacts/sosre
 ```
 Where `${TEST_NAME}` is the test name directory (e.g., `e2e-aws-tests`, `e2e-aws-ovn-ocp-conformance-serial`). Use `find ${TMP}/artifacts -name 'sosreport-*.tar.xz'` to locate it.
 
+## Work Directory
+
+Set once at the start and reference throughout:
+```bash
+WORKDIR=/tmp/analyze-ci-claude-workdir.$(date +%y%m%d)
+mkdir -p ${WORKDIR}
+```
+
 ## Common Commands
 
 Create a temporary working directory to store artifacts for the current job:
 ```bash
-mkdir -p /tmp/analyze-ci-claude-workdir
-mktemp -d /tmp/analyze-ci-claude-workdir/openshift-ci-analysis-XXXX
+mktemp -d ${WORKDIR}/openshift-ci-analysis-XXXX
 ```
 
 Fetch the high level summary of the failed prow job:

.claude/commands/analyze-ci-for-pull-requests.md

@@ -20,11 +20,18 @@ This command orchestrates the analysis workflow by:
 1. Fetching the list of open PRs and their failed jobs using `.claude/scripts/microshift-prow-jobs-for-pull-requests.sh --mode detail`
 2. Filtering to only PRs that have at least one failed job
 3. Analyzing each failed job individually using the `/analyze-ci-for-prow-job` command
-4. Aggregating results into a summary report saved to `/tmp/analyze-ci-claude-workdir`
+4. Aggregating results into a summary report saved to `${WORKDIR}`
 
 ## Arguments
 - `--rebase` (optional): Only analyze rebase PRs (titles containing `NO-ISSUE: rebase-release-`)
 
+## Work Directory
+
+Set once at the start and reference throughout:
+```bash
+WORKDIR=/tmp/analyze-ci-claude-workdir.$(date +%y%m%d)
+```
+
 ## Implementation Steps
 
 ### Step 1: Fetch Open PRs and Their Job Results
@@ -76,13 +83,13 @@ Each job MUST be analyzed by launching a **separate Agent** (using the `Agent` t
 3. The file MUST contain the full report including the `--- STRUCTURED SUMMARY ---` block
 
 **Actions**:
-1. Run `mkdir -p /tmp/analyze-ci-claude-workdir` using the `Bash` tool
+1. Run `WORKDIR=/tmp/analyze-ci-claude-workdir.$(date +%y%m%d) && mkdir -p ${WORKDIR}` using the `Bash` tool
 2. For each failed job URL, launch a separate **Agent** with this exact prompt template:
    ```
    Agent: subagent_type=general_purpose, prompt="Analyze this Prow job and save the report:
    1. Run /analyze-ci-for-prow-job <JOB_URL>
    2. After the analysis completes, save the FULL report output (including the --- STRUCTURED SUMMARY --- block) to:
-      /tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-<N>-pr<PR>-<JOB_NAME_SUFFIX>.txt
+      ${WORKDIR}/analyze-ci-prs-job-<N>-pr<PR>-<JOB_NAME_SUFFIX>.txt
       Use the Write tool to save the file. The file must contain the complete analysis report."
    ```
    Replace `<JOB_URL>`, `<N>` (1-based job index), `<PR>` (PR number), and `<JOB_NAME_SUFFIX>` (last segment of job name) with actual values.
@@ -108,7 +115,7 @@ For each job analysis, extract:
 
 **File Storage**:
 All per-job report files MUST be saved at this exact path:
-- `/tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-<N>-pr<PR>-<job-name-suffix>.txt`
+- `${WORKDIR}/analyze-ci-prs-job-<N>-pr<PR>-<job-name-suffix>.txt`
 
 Where:
 - `<N>` is the 1-based job index
@@ -117,7 +124,7 @@ Where:
 
 **Verification**: After all agents complete, verify that per-job files exist:
 ```bash
-ls /tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-*.txt
+ls ${WORKDIR}/analyze-ci-prs-job-*.txt
 ```
 If any files are missing, note the gap in the summary report but do NOT re-run the analysis.
 
@@ -127,7 +134,7 @@ If any files are missing, note the gap in the summary report but do NOT re-run t
 
 **Actions**:
 1. Collect results from all parallel job analyses
-   - Read each per-job file from `/tmp/analyze-ci-claude-workdir`
+   - Read each per-job file from `${WORKDIR}`
    - Extract the `--- STRUCTURED SUMMARY ---` block from each file for pattern detection (including the `FINISHED` field for job date tracking)
    - Extract key findings from each analysis
 
@@ -145,7 +152,7 @@ If any files are missing, note the gap in the summary report but do NOT re-run t
 **Actions**:
 1. Aggregate all job analysis results from parallel execution
 2. Identify common patterns and group by PR and failure type
-3. Generate summary report and save to `/tmp/analyze-ci-claude-workdir/analyze-ci-prs-summary.<timestamp>.txt`
+3. Generate summary report and save to `${WORKDIR}/analyze-ci-prs-summary.<timestamp>.txt`
 4. Display the summary to the user
 
 **Important**: Each failed job MUST include the finish date in `[YYYY-MM-DD]` format (from the per-job `FINISHED` field) after the job name. This ensures the HTML report generator can extract dates without reading per-job files.
@@ -162,7 +169,7 @@ OVERVIEW
   PRs with Failures: 2
   Total Failed Jobs: 9
   Analysis Date: 2026-03-15
-  Report: /tmp/analyze-ci-claude-workdir/analyze-ci-prs-summary.20260315-143022.txt
+  Report: ${WORKDIR}/analyze-ci-prs-summary.20260315-143022.txt
 
 PER-PR BREAKDOWN
 
@@ -199,7 +206,7 @@ COMMON PATTERNS (across PRs)
 
 ═══════════════════════════════════════════════════════════════
 
-Individual job reports: /tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-*.txt
+Individual job reports: ${WORKDIR}/analyze-ci-prs-job-*.txt
 ```
 
 ## Examples
@@ -231,7 +238,7 @@ Individual job reports: /tmp/analyze-ci-claude-workdir/analyze-ci-prs-job-*.txt
 - **Execution Time**: Depends on number of failed jobs; parallel execution helps significantly
 - **Network Usage**: Each job analysis fetches logs from GCS
 - **Parallelization**: All job analyses run in parallel for maximum efficiency
-- **File Storage**: All intermediate and report files are stored in `/tmp/analyze-ci-claude-workdir` directory
+- **File Storage**: All intermediate and report files are stored in `${WORKDIR}` directory
 
 ## Prerequisites
 
@@ -272,7 +279,7 @@ Please ensure you're in the microshift project directory.
 
 - This skill focuses on **presubmit** PR jobs (not periodic/postsubmit)
 - Analysis is read-only - no modifications to CI data or PRs
-- Results are saved in files in /tmp/analyze-ci-claude-workdir directory with a timestamp
+- Results are saved in files in ${WORKDIR} directory with a timestamp
 - Provide links to the jobs in the summary
 - Only present a concise analysis summary for each job
 - PRs with no Prow jobs (e.g., drafts without triggered tests) are skipped

.claude/commands/analyze-ci-for-release-manager.md

@@ -18,15 +18,23 @@ Accepts a comma-separated list of MicroShift release versions, runs the `analyze
 ## Arguments
 - `$ARGUMENTS` (required): Comma-separated list of release versions (e.g., `4.19,4.20,4.21,4.22`)
 
+## Work Directory
+
+Set once at the start and reference throughout:
+```bash
+WORKDIR=/tmp/analyze-ci-claude-workdir.$(date +%y%m%d)
+```
+
 ## Implementation Steps
 
 ### Step 1: Parse and Validate Arguments
 
 **Actions**:
-1. Split `$ARGUMENTS` by comma to get a list of release versions
-2. Trim whitespace from each version
-3. Validate that at least one release version is provided
-4. If no arguments provided, show usage and stop
+1. Run `WORKDIR=/tmp/analyze-ci-claude-workdir.$(date +%y%m%d) && mkdir -p ${WORKDIR}` using the `Bash` tool
+2. Split `$ARGUMENTS` by comma to get a list of release versions
+3. Trim whitespace from each version
+4. Validate that at least one release version is provided
+5. If no arguments provided, show usage and stop
 
 **Error Handling**:
 - If `$ARGUMENTS` is empty, display: "Usage: /analyze-ci-for-release-manager <release1,release2,...>" and stop
@@ -39,7 +47,7 @@ Accepts a comma-separated list of MicroShift release versions, runs the `analyze
    Agent: subagent_type=general_purpose, prompt="Run /analyze-ci-for-release <version>"
    ```
 2. Launch all releases **in parallel** as separate agents — do NOT wait for one to finish before starting the next
-3. After each agent completes, note the summary report file path it produced (typically `/tmp/analyze-ci-claude-workdir/analyze-ci-release-<version>-summary.*.txt`)
+3. After each agent completes, note the summary report file path it produced (typically `${WORKDIR}/analyze-ci-release-<version>-summary.*.txt`)
 4. Wait until all the parallel agents are complete
 5. Track which releases succeeded and which failed
 
@@ -56,7 +64,7 @@ Analyzing release X/Y: <version>
    Agent: subagent_type=general_purpose, prompt="Run /analyze-ci-for-pull-requests --rebase"
    ```
 2. This agent can be launched in parallel with the release agents in Step 2
-3. After the agent completes, note the summary report file path (typically `/tmp/analyze-ci-claude-workdir/analyze-ci-prs-summary.*.txt`)
+3. After the agent completes, note the summary report file path (typically `${WORKDIR}/analyze-ci-prs-summary.*.txt`)
 4. If no rebase PRs are found, note "No open rebase PRs" for the report
 
 **Progress Reporting**:
@@ -93,7 +101,7 @@ Summary:
   Pull Requests:
     2 rebase PRs with 5 total failed jobs
 
-HTML report generated: /tmp/analyze-ci-claude-workdir/microshift-ci-release-manager-20260315-143022.html
+HTML report generated: ${WORKDIR}/microshift-ci-release-manager-20260315-143022.html
 ```
 
 ## Examples
@@ -136,7 +144,7 @@ HTML report generated: /tmp/analyze-ci-claude-workdir/microshift-ci-release-mana
 - All agents (releases + PR analysis) are launched in parallel for maximum efficiency
 - HTML generation is delegated to `analyze-ci-generate-html-report` running as a separate agent — it only reads summary files (not per-job files), keeping context usage minimal
 - The HTML report is self-contained (no external CSS/JS dependencies)
-- All intermediate files from `analyze-ci-for-release` and `analyze-ci-for-pull-requests` remain available in `/tmp/analyze-ci-claude-workdir`
+- All intermediate files from `analyze-ci-for-release` and `analyze-ci-for-pull-requests` remain available in `${WORKDIR}`
 - The HTML file can be opened in any browser for convenient examination
 - If a release analysis fails, it is noted in the report but does not block other releases
 - If no rebase PRs are open, the Pull Requests tab shows "No open rebase pull requests found"