sergeyklay/.agents / .agents/skills/improve-self
improve-self
VerifiedDetect when a missing Agent Skill would have made the current task faster or more reliable, then scaffold a candidate SKILL.md for user review. Use after a non-trivial task when one of these signals fires: the same procedure repeated 3+ times, error recovery required inventing a workflow, the user corrected the approach (not the answer), the agent realized mid-task that a playbook did not exist, or 5+ tool calls produced what should have taken 1-2. Detects which `.{vendor}/` directories exist on the host (project and home), reads every skill and slash command in them before proposing, and never duplicates. Writes candidates into the project's .{vendor}/skills/{name} directory that the running host actually loads, never the home folder. Do NOT use for first-attempt failures, one-off questions, fact gaps that belong in research, or style gaps that belong in rules files. Pass the trace and the signal that fired as arguments; the skill does not reconstruct them from the conversation.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 15.2 KB |
| references/alternatives-to-a-skill.md | markdown | 1.2 KB |
| references/discover-skills-reference.md | markdown | 4.6 KB |
| references/platform-vendors.md | markdown | 1.9 KB |
| scripts/discover_skills.pyexecutable | script | 19.7 KB |
references/platform-vendors.md
View on GitHub# Platform → vendor mapping
Each agent platform loads skills from a known set of `.{vendor}/skills/` paths. Use the running agent's self-knowledge — its system prompt, its `AGENTS.md` context file, or its platform documentation — to pick the vendor names below. Do not equate "exists on disk" with "I will load this".
| Running agent | Vendor names to pass |
|---|---|
| Anthropic Claude Code | `claude`, plus `agents` for discovery only: Claude Code never loads `.agents/skills/` |
| OpenAI Codex | `agents` |
| Cursor | `cursor` (also `agents` if configured to read the cross-platform convention) |
| Gemini CLI | `gemini`, `agents` |
| GitHub Copilot (VS Code) | `github` (project scope), `copilot` (user scope) |
| OpenCode / Antigravity | `agents` |
| Windsurf | `windsurf` |
Treat the table as a reference, not a permanent contract — vendor lists drift as platforms adopt or drop the cross-platform convention.
## Degraded fallback: filesystem probe
If the agent cannot determine its own identity from context (rare, but possible inside sandboxed sub-agents), probe the filesystem and mark the resulting list as uncertain in the response:
```bash
detected=$({
find . -maxdepth 2 -type d -path './.*/skills' 2>/dev/null
find "${HOME}" -maxdepth 2 -type d -path "${HOME}/.*/skills" 2>/dev/null
} | awk -F/ '{name=$(NF-1); sub(/^\./, "", name); print name}' \
| sort -u | paste -sd, -)
echo "detected vendors (uncertain): ${detected:-<none>}"
```
The filesystem probe **over-reads**: it returns every vendor prefix present on disk, including ones the running agent does not actually load. Prefer the self-knowledge table above whenever it applies.
## Empty result
If the resolved list is empty either way (no readable vendor directories), skip the Phase 2 script call and treat the result as "no existing skills to deduplicate against" — the candidate gap proceeds to Phase 3 unimpeded.
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: 472k | 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 |