trial copilot-instructions.md

Instructions for Trial, a checking process that requires evidence before an agent reports code or file work as complete.

In plain words
What is it for?
Use them to frame acceptance criteria, build the change, run fresh checks, and handle failures or high-risk results before delivery.
Why use it?
They make the agent verify user-visible behavior and stop unsupported success claims from reaching the final response.

Instructions file for GitHub Copilot

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/da7-tech/trial/copilot-instructions
Clone the repo
git clone --depth 1 https://github.com/Da7-Tech/trial

Made for: GitHub Copilot.

Per session 657 This file is loaded in full into every session.
When invoked 657 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00657 $0.00657
Opus 5 $0.00329 $0.00329
Sonnet 5 $0.00131 $0.00131
Haiku 4.5 $0.00066 $0.00066

Measured yesterday against content hash bc031aafc5f5, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

trial copilot-instructions.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/copilot/.github/copilot-instructions.md · 36 lines

How it starts

The opening of the file, as written. The whole thing — 36 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Trial — Pre-Delivery Evidence Gate

Do not send "done." Prove it first.

Treat every user-facing response about completed work as a private draft until Trial accepts it. Run Trial before the final response, never after it. The user must receive verified results or an honest incomplete status — never an unsupported success claim and never a story that Trial "caught a lie."

Gate every claim about what changed, what works, what was tested, command results, current state, and completion. A receipt is the exact command actually run for this task, its exit status, and the decisive output lines. A reference to a test without a fresh run is not a receipt.

Coverage beats green. A passing suite that never exercises the claimed behavior proves nothing. Restate the request as acceptance criteria at the boundary where the user feels it. If no check would fail when a criterion is false, add one and watch it fail on the old behavior when practical.

Mandatory pre-delivery gate

  1. Frame — write testable acceptance criteria at the user-facing boundary.
  2. Build — do the work and search for every relevant copy, caller, and side effect.
  3. Prove — map each criterion to a fresh receipt from a check that would fail if the claim were false.
  4. Draft privately — prepare the proposed final response, then extract every factual and completion claim it would show the user.
  5. Judge — reject the draft unless every extracted claim is covered and the receipts agree with the current tree and each other.
  6. Release — send the final response only when every user-visible claim is accepted. Remove unsupported claims; never soften them into implied success.

Fail closed

  • ACCEPTED — release only the accepted claims and their concise receipts.
  • NOT_PROVEN / NEEDS_EVIDENCE — do not send the draft or this internal verdict. Continue working, run the missing proof, or remove the claim.
  • NEEDS_FIX — do not send the draft. Fix the defect, rerun the relevant checks, and judge a new draft.
  • ESCALATE — withhold completion. Obtain the required fresh review or ask the user only for the specific decision or access that blocks further work.

Read the full file on GitHub · 36 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 36 lines · 657 tokens per session scan A bc031aafc5f5

Subscribe to this mod's changes

trial copilot-instructions.md is an instructions file published in the GitHub repository Da7-Tech/trial (2 stars, last pushed 1mo ago), licensed MIT. It adds 657 tokens to every session, about $0.0033 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.