technical-skill-finder
Mine coding agent logs (Codex/Cursor/session histories and similar telemetry) to discover high-value candidate skills, then draft structured skill creation/reuse recommendations.
How do I install this agent skill?
npx skills add https://github.com/vincentkoc/dotskills --skill technical-skill-finderIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill mines local developer agent logs to identify patterns for new technical skills. While it accesses sensitive history files, it includes explicit privacy requirements and guardrails to minimize data exposure.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
- Runlayerwarn
4/4 files flagged
What does this agent skill do?
Technical Skill Finder
Purpose
Find recurring pain points from local agent logs and convert them into actionable skill candidates, reuse opportunities, or existing skill updates.
When to use
- You want to discover missing technical skills from historical agent activity.
- You want reproducible criteria before creating a new skill.
- You want to validate whether an existing skill already covers the pattern.
- You want to include optional personal-signal sources (when authorized).
Inputs
SCOPE(required): repository paths, workspace, or tool domains to inspect.SOURCES(required): ordered source list to mine.TIMEFRAME(optional): defaultallunless constrained by user.PRIVACY_POLICY(required): explicit user direction for personal logs.TOP_N(optional): number of highest-priority candidates to return.
Workflow
- Initialize source set
~/.codex/history.jsonl~/.codex/archived_sessions/*.jsonl~/.codex/sessions/*.jsonland~/.codex/log/*if present- Repository-specific telemetry in
AGENTS.md/local docs when available Cursor/Codexagent logs detected under known dotfiles directories
- Normalize extraction signals
- Parse stack traces and classify failure type (
auth,type-check,llm-error,git/ci,runtime,refactor-merge,test) - Parse recurring command phrases (
rg,mypy,pytest,gh,git, package-manager failures) - Record frequency, recency, and affected project context
- Parse stack traces and classify failure type (
- Cluster signals
- Group by: domain (python/js/rust/docs/tooling), command lineage, and error signature.
- Deprioritize one-off sessions with low recurrence.
- Map to existing skills
- Compare candidate clusters with available skills by
nameanddescription. - If overlap is high, propose skill update path.
- If no overlap, propose new skill.
- Compare candidate clusters with available skills by
- Emit ranking output
- Provide
impact,frequency,confidence,skill-fit, and first-apply command set.
- Provide
- Produce minimal first-iteration output for high-priority candidates
- Candidate title + scope
- Trigger phrase examples
- Required inputs
- Suggested workflow summary
- Evidence snippets (line/file-level)
- Suggested dependencies/tools (e.g.,
jq,rg, shell utilities, MCP resources) - Return this through chat/stdout by default. Create a persistent artifact root only when the user selects one or another required workflow declares it, with file/byte budgets and source/input identity.
- Optional extension to personal-signal sources
- Only after explicit approval to read personal channels.
- If MCP is available and user has granted access, run MCP resource discovery and include message-signal-derived patterns.
- Keep this opt-in and isolated from coding-signal output unless user requests a merged plan.
Guardrails
- Never infer or emit private content from message logs unless explicitly permitted.
- Skip binary/corrupt files and summarize only parseable text sources.
- Prefer deterministic commands and small scripts over ad-hoc manual parsing.
- Always avoid proposing skills with unresolved operational context (credentials, environment, private URLs).
- If evidence is ambiguous, return
confidence: lowand request one more session sample. - Reuse one canonical identity-matched inventory instead of materializing
duplicate large extracts. Apply the
$operations-worktreetask artifact contract when retention or resumable phase state is required.
Outputs
skill_candidates.md-style report in chat:reusecandidates (existing skill can be extended)newskill candidates (not yet covered)- top source anchors with references
- recommended next action (create/update)
Read references/sources.md for source precedence.
Read references/scorecard.md for prioritization rules.
Flow
stateDiagram-v2
[*] --> SelectAuthorizedSources
SelectAuthorizedSources --> NormalizeAndCluster
NormalizeAndCluster --> CompareExistingSkills
CompareExistingSkills --> ProposeUpdate: substantial overlap
CompareExistingSkills --> ProposeNewSkill: no existing coverage
CompareExistingSkills --> ReportUncertainty: insufficient evidence
ProposeUpdate --> RankAndReport
ProposeNewSkill --> RankAndReport
RankAndReport --> [*]
ReportUncertainty --> [*]
note right of SelectAuthorizedSources
Personal channels require explicit approval.
Return findings inline unless retention is selected.
end note
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/vincentkoc/dotskills/technical-skill-finder">View technical-skill-finder on skillZs</a>