Agent Skills

Reusable skills and agents for AI coding agents, primarily Claude Code.

Why ora

Most Claude Code agent frameworks (11–24 agents, 9+ hooks) add complexity to compensate for weaker models. With Opus 4.6, that complexity burns tokens without improving output. ora ships exactly two research agents:

• Ariadne — codebase exploration (enhanced contextual grep over local files)
• Clio — external research (docs, web, GitHub repos)

Both isolate search context from the main conversation — broad queries never pollute your main window. The plugin has no hooks and no planning/verification/execution agents. Planning and verification happen inline in the main agent, shaped by behavioral rules in your CLAUDE.md (see Configuration below).

Getting Started

Install

# Plugin (agents)
/plugin marketplace add trancong12102/agentskills
/plugin install ora@agentskills

# Skills (optional, standalone)
bunx skills add trancong12102/agentskills -g -y -a claude-code

# Other plugins
/plugin install sound-notify@agentskills

Credentials

[profiles.oracle]
model = "gpt-5.4"
model_reasoning_effort = "xhigh"
approval_policy = "never"
sandbox_mode = "read-only"

ora Plugin

Configuration

ora is just two agents — workflow behavior lives in your ~/.claude/CLAUDE.md. Recommended setup:

<investigate_before_responding>
Do not respond to tech task without loading matching skill or spawning ora:Clio research first. Cost near zero; stale knowledge break implementations. Skip only when task clearly unrelated to any skill.

Never speculate on code not opened. User reference specific file → read file first. Ground all claims in investigated state.
</investigate_before_responding>

<subagent_routing>
Do not use built-in Explore or general-purpose agent — use ora:Ariadne (local codebase) and ora:Clio (external sources). general-purpose escape hatch only when no ora agent fits.

Do not serialize independent subagent work. Fan out parallel: spawn multiple subagents in single turn when work covers independent items/files.

Do not respawn agents. Follow-up on agent already spawned this session → SendMessage with returned agentId. Fresh spawn lose cached context and repeat tool calls. New Agent call only when topic genuinely different.
</subagent_routing>

<plan_before_implementing>
Do not implement without calling EnterPlanMode tool first when task ambiguous, spans multiple subsystems, or acceptance criteria unclear. Skip plan mode for clearly-scoped changes: single-file bug fix with obvious fix site, mechanical rename/refactor, config/typo fix.
</plan_before_implementing>

<think_before_coding>
Do not start implementing without stating assumptions and flagging tradeoffs. Multiple reasonable interpretations → present them, do not pick silent. Simpler approach exist than asked → say so, push back. Unclear enough to block correct execution → stop and ask, do not guess.
</think_before_coding>

<goal_driven_execution>
Do not accept vague goals. Translate each task into verifiable success criterion before implementing ("add validation" → "write tests for invalid inputs, then make them pass"). Do not mark task complete until success criterion met — read actual code, run actual check, do not trust own summary.
</goal_driven_execution>

<ask_user_question_tool>
Do not ask user via plain response. Use AskUserQuestion tool — load via `ToolSearch` with `select:AskUserQuestion`.
</ask_user_question_tool>

think_before_coding and goal_driven_execution are adapted from forrestchang/andrej-karpathy-skills.

License

MIT — Cong Tran