Bootstrap

bootstrap — session enabler

The only bootstrap-invoked skill. Injected by the adapter at session start and after compaction. Teaches two habits: (1) check for and invoke relevant skills before any response, and (2) check for project artifacts (CONTEXT.md, active plan folder) and keep them updated.

This is the skill that makes the system self-triggering. Without it, the other skills sit unused.

WHEN (bootstrap-invoked)

Injected at:

• session_start — a new session begins.
• session_compact — context was compacted; re-establish the habits.

Not user-invoked. Not model-invoked. Only the adapter injects it.

HABIT 1 — CHECK FOR AND INVOKE RELEVANT SKILLS

Before any response (including clarifying questions), check whether a skill might apply. If there is even a small chance a skill applies, invoke it to check. If an invoked skill turns out to be wrong for the situation, you don't have to use it.

user message received
  -> might any skill apply?
       yes (even a small chance) -> invoke the skill; follow it exactly
       definitely not           -> respond

How to invoke a skill

Use the harness's skill-loading mechanism (the adapter maps this). Never read skill files manually with file tools when a skill-loading mechanism exists — the mechanism activates the skill properly.

Instruction priority

1. User's explicit instructions (AGENTS.md, direct requests) — highest.
2. Skills — override default system behavior where they conflict.
3. Default system prompt — lowest.

If AGENTS.md says "don't use TDD" and the tdd skill says "always use TDD", follow the user's instructions. The user is in control.

When invoked as a subagent

If you were dispatched as a subagent to execute a specific task, skip the skill-checking habit — your dispatcher already constructed your context. Focus on the task.

HABIT 2 — CHECK FOR PROJECT ARTIFACTS

After the skill check, check the project state:

1. CONTEXT.md at repo root — if it exists, read it for canonical terms. Use those terms throughout. If a term is fuzzy, invoke the domain-modeling discipline.
2. Active plan folder — resolve per AGENTS.md §6:
   • User named it: build docs/plans/2026-06-28-fix-auth-timeout.
   • Conversation context: the plan slug matches what the user is discussing.
   • Most-recently-modified plan folder, if still ambiguous.
   • Ask, if still ambiguous. If an active plan exists, read its artifacts (PRD.md, ARD.md, SPEC.md, PLAN.md, LESSONS.md, RESEARCH.md) before responding to plan-related work.

The write habit

When something important happens, write it to the appropriate artifact:

• A new term is coined -> update CONTEXT.md.
• A requirement is clarified -> update PRD.md in the active plan.
• An invariant is discovered -> update ARD.md or SPEC.md.
• A bug is fixed -> append to LESSONS.md.

The artifacts are NOT memory — they are externalized state the agent maintains. The habit of checking and updating them is what this skill teaches. (See AGENTS.md §10.)

ANNOUNCE

When you invoke a skill as a result of the habit, announce briefly: "Using the <skill> skill to <purpose>." This keeps the user oriented without noise.

BOUNDARIES

• MUST check for relevant skills before any response (even clarifying questions).
• MUST check for CONTEXT.md and an active plan folder.
• MUST use canonical terms from CONTEXT.md.
• MUST follow user instructions over skill defaults when they conflict.
• MUST NOT invoke skills when dispatched as a subagent for a specific task.
• MUST NOT write artifacts during the check — only read. Other skills write.
• MUST NOT be user-invoked or model-invoked. Only the adapter injects it.