feat: AI-friendly structured workspace for Claude Code / Codex

[STRRL]

Summary

• Replace all old CLI commands (analyze, debug *) with a new workspace command group: create, add-log, analyze
• Build structured file-based workspaces with pattern directories named by LLM-generated semantic IDs — navigable by any AI coding agent or human using basic unix tools
• New pkg/workspace/ package with embedded Go templates for AGENTS.md, pattern.md, summary.md, errors.md
• Refactor pkg/analyzer/ to expose RunAgentWithPrompt() and BuildWorkspaceSystemPrompt() for the new workspace layout
• Remove DuckDB dependency from CLI (--db flag removed); pkg/store/ kept for tests/future use

New CLI

lapp workspace create <dir>
lapp workspace add-log <dir> <logfile> [--model] [--stdin]
lapp workspace analyze <dir> [question] [--model]

Workspace structure

my-investigation/
├── AGENTS.md
├── logs/
├── patterns/
│   ├── server-startup/
│   │   ├── pattern.md
│   │   └── samples.log
│   ├── connection-timeout/
│   │   ├── pattern.md
│   │   └── samples.log
│   └── unmatched/
│       └── samples.log
└── notes/
    ├── summary.md
    └── errors.md

Closes #17

Test plan

• make ci passes (fmt, vet, build, lint, unit tests)
• lapp workspace create /tmp/test-ws creates directory structure with AGENTS.md
• lapp workspace add-log /tmp/test-ws <logfile> populates patterns/ and notes/
• echo "test" | lapp workspace add-log /tmp/test-ws --stdin reads from stdin
• lapp workspace analyze /tmp/test-ws "what errors?" runs agent

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f05c01eb0d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve templates when semantic labels are missing

computePatterns drops any Drain template not present in the LLM label map (if !hasLabel { continue }), but semantic.parseResponse accepts partial arrays and does not verify one label per input pattern. When the model omits one pattern_id, all lines for that template are later treated as unmatched, so patterns/ and notes/summary.md lose real clusters and counts become incorrect. Keep unlabeled templates with a fallback semantic ID/description instead of skipping them.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Avoid overwriting existing logs with same basename

add-log always writes to logs/<basename> with os.WriteFile, so adding two different files that share the same filename (for example /var/log/app.log then /tmp/archive/app.log) silently replaces the first file. Because rebuilds process whatever is in logs/, this can drop previously ingested data and change pattern discovery unexpectedly. Use a collision-safe destination name (or fail on conflict) instead of unconditional overwrite.

Useful? React with 👍 / 👎.