piece
A part you drop into a harness.
- Skill — A named procedure an agent follows when a situation matches.
- Prompt — A single instruction, written for one job.
- Rule — A standing instruction the model is expected to keep.
- Token zero — The short text that is present at the start of every turn.
- Hook — A small program that runs at a named moment in the harness.
- Context injection — A named slot that inserts data into the window. The slot is not authority.
- Subagent — A bounded helper with a job, tools, and a stop condition.
- Bot — A standing clerk with a queue and no authority past that queue.
- Workflow — An ordered path from a trigger to a result.
- Goal — The outcome that tells you the work is finished.
- Context window — One engineered window, one-shot or multi-shot, with a boundary.
Published
A boot note is data
Context injectionpieceCC-BY-4.0
A named slot that inserts a boot note into the window between markers. The note is data inside a boundary. It does not outrank the seat.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
Defect pass
PromptpieceCC-BY-4.0
A prompt that judges one piece of work against a written checklist, lists defects, and stops. It does not rewrite the work it was asked to judge.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
How to file a specimen
SkillpieceCC-BY-4.0
A skill an agent follows when asked to file one reusable artifact on the public shelf. It names the situation, the steps, and a refusal. It is a procedure, not …
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
Intake clerk
BotpieceCC-BY-4.0
A standing clerk that accepts a submission into a pending queue or refuses it. Approval is a different role. The clerk cannot approve.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
One specimen in the window
Context windowpieceCC-BY-4.0
A one-shot window that holds exactly one specimen and one question: what you would change to reuse it, and what you would leave.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
Read-only verifier
SubagentpieceCC-BY-4.0
A bounded helper that reads one artifact, runs the check that artifact names, and reports pass or fail. It has no write tools and no publish tools.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
Record, do not publish
HookpieceCC-BY-4.0
A small end-of-turn program that records that the turn ended. It refuses to publish, send, delete, or decide that the work succeeded.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
A short public seat
Token zeropieceCC-BY-4.0
Sixteen lines that are present at the start of every turn: one home for each fact, verify before a claim, and ask before a destructive act. Public teaching text…
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
A stranger reuses one
GoalpieceCC-BY-4.0
The finish line for a single specimen: someone who does not know the author can reuse it, without asking a question the specimen should have answered.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
Submit, moderate, read
WorkflowpieceCC-BY-4.0
The library spine as an ordered path: a person submits, a person publishes, and a stranger reads and copies. A sensor may advise the middle step and cannot publ…
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.
Verify before you claim
RulepieceCC-BY-4.0
A standing rule: a claim that work is done, fixed, or published is allowed only when the speaker can point at an observation a stranger can rerun.
Original teaching specimen written for the Campaign Hub library. Not a transcript of a private run.