leek/agent-skills / skills/retro
retro
UnverifiedLook back over AI-coding sessions (this one, or recent session logs) and propose improvements to the agent's environment.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.9 KB |
| agents/openai.yaml | config | 187 B |
| references/session-selection.md | markdown | 2.5 KB |
| scripts/extract-signal.shexecutable | script | 7.4 KB |
| scripts/select-sessions.shexecutable | script | 15.7 KB |
references/session-selection.md
View on GitHub# Session selection
`scripts/select-sessions.sh` does discovery, grouping, ranking, and batching. This file
holds what it cannot decide: who wrote a session.
## What the script already does
- Pre-filters each harness to its newest files by mtime, so old archives are never
streamed, then ranks roots by the conversation's own timestamp, not the file's.
- **Claude:** `<uuid>.jsonl` owns `<uuid>/subagents/agent-*.jsonl`; harness agent
worktrees are skipped.
- **Codex:** identity from the first `session_meta` (`id`, else `session_id`), a plain
and compressed copy deduplicated; a child (`thread_source: subagent` or a
`source.subagent` object) is followed up its `parent_thread_id` chain to its root. A
child whose root is not available is excluded as an orphan.
- **Grok:** `summary.json` gives the kind (`subagent*` children attach to the root whose
history names their id) and the last-active time; `<user_query>` holds the human text.
- **agy:** `history.jsonl` joins `conversationId` to the `.db` basename.
- Hints that exclude a root unless `--keep-automated`: a headless entrypoint (`sdk-cli`,
`sdk-py`, `codex exec`, Grok `headless`), a panelist brief (for agy, also one found in
the step 14 prompt), a Claude session opened by a relayed teammate message, an agy
conversation absent from history with no step 14 prompt in its `.db` (relayed or headless). A `/private/tmp`
cwd, or an agy prompt found only as step 14, sets `clue`.
- Roots rank by their latest activity, linked children included.
## Decide authorship
Read each selected root's `first_prompt`, and, when that is ambiguous, its first few
`user` events from `extract-signal.sh`.
- A `/private/tmp` cwd or a panelist-shaped brief is a clue, not proof; classify from the
actual task. Keep an automated panelist's log as linked evidence when its human parent
is identifiable.
- An agy prompt that is not in history is step 14 or a relayed step 101 in the `.db`. A
relayed prompt is not another human correction.
- Prompt history can miss queued corrections; the transcript's `queued` events hold them.
- Mark unresolved authorship as unknown and exclude it from claims about human
correction frequency.
- A root that is not a human session: rerun with `--exclude <id>`, so the count stays N.
## Report
Selected roots by harness, linked children, exclusions by reason (automated, orphan,
unknown, caller), unreadable files, and missing harnesses. The batch files and these
totals must agree before synthesis.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |