npx agentspec-cli install 7da590df-68ca-4876-9fa8-08603639927d --file-type copilot-instructions→ .github/instructions/
poml<spec agent="github-copilot" version="1.0">
<pre_flight>
<gather>
<item>Goal and constraints (perf, security, deadlines)</item>
<item>Repo structure and entry points</item>
<item>Existing reusable code (prefer extend over rewrite)</item>
<item>Default language: python (.py) unless specified</item>
</gather>
</pre_flight>
<task>
<phase name="plan">
<step>Produce a 3–5 step checklist WITH acceptance criteria.</step>
<step>Audit repo for reuse opportunities and dependencies.</step>
<step>Identify risks and required confirmations.</step>
</phase>
<phase name="execute">
<step>Mark checklist progress: [ ] → [x].</step>
<step>Write/modify code in project files (not chat).</step>
<step>Show minimal git diff preview before commit.</step>
<step>Commit atomic changes with clear messages.</step>
</phase>
<phase name="verify">
<step>Run tests, lint, type checks, and build if relevant.</step>
<step>Report results succinctly.</step>
<step>Declare DONE or provide next TODO step.</step>
</phase>
</task>
<completion_criteria>
<rule>MVP first, iterate after feedback.</rule>
<rule>All tests/checks pass and diff reviewed.</rule>
<rule>Code stored in .py files by default.</rule>
<rule>Acceptance criteria met.</rule>
</completion_criteria>
<guardrails>
<security>
<rule>Never commit secrets/tokens.</rule>
<rule>Ensure .gitignore covers credentials and artifacts.</rule>
<rule>Use environment variables or secret stores.</rule>
</security>
<quality>
<rule>Add/extend tests alongside code changes.</rule>
<rule>Run lint and type checks; fix blockers.</rule>
<rule>Pin dependency versions (lockfiles or exact pins).</rule>
</quality>
<risk_management>
<rule>No new dependencies without permission.</rule>
<rule>Confirm destructive ops (rm, force-push, schema changes).</rule>
<rule>Provide explicit rollback steps for risky changes.</rule>
</risk_management>
</guardrails>
<terminal_protocol>
<proposal>
<action></action> <!-- exact command -->
<purpose></purpose> <!-- why this is needed -->
<risk>low|medium|high</risk>
<undo></undo> <!-- rollback command/steps -->
</proposal>
<execution>
<rule>Await CONFIRM for any destructive action.</rule>
<rule>Log outcomes and update checklist state.</rule>
</execution>
</terminal_protocol>
<complexity>
<definition>
<rule>Touches ≥2 modules/services OR introduces a new service/DB/SDK.</rule>
<rule>Includes migrations, infra changes, or other destructive ops.</rule>
<rule>Requirements are ambiguous or require external coordination.</rule>
<rule>Estimated effort > 60 minutes OR > 50 LOC changed across files.</rule>
</definition>
<is_complex>TRUE if any rule matches; otherwise FALSE.</is_complex>
<report required_when="TRUE" name_format="reports/{YYYYMMDD-HHmmss}_{report_name}.md" timezone="local">
<must_include>
<item>Goal, scope, constraints, assumptions.</item>
<item>Architecture/flow (ASCII diagram acceptable).</item>
<item>Plan checklist with owners and timestamps.</item>
<item>Risk analysis, confirmation points, rollback steps.</item>
<item>Links to diffs, tests, artifacts, and follow-ups.</item>
</must_include>
</report>
</complexity>
<response_templates>
<simple_task>
PLAN:
- [ ] Define acceptance criteria (MVP first)
- [ ] Check reusable code and deps
- [ ] Implement in .py file
- [ ] Add/extend tests
- [ ] Run lint/type/tests
FILES: <path>:<summary_of_changes>
DIFF: <minimal_preview>
CHECK: <test_lint_build_results>
STATUS: DONE | TODO: <next_step>
</simple_task>
<error_handling>
ATTEMPT: <what_was_tried>
ERROR: <short_summary>
CAUSE: <likely_reason>
FIX: <one_or_two_options>
REQUEST: CONFIRM option or provide more context
</error_handling>
</response_templates>
<optimization_rules>
<rule>Prefer the smallest viable change that meets DoD.</rule>
<rule>Compose over inherit; prefer pure functions over stateful.</rule>
<rule>Ask at most one clarification; otherwise present two concrete paths.</rule>
<rule>Stop after two failed attempts and request direction.</rule>
</optimization_rules>
<allowed_commands>
<safe>mv, cp, mkdir, touch, git add, git commit</safe>
<confirm_required>rm, git reset --hard, force-push, db migrations</confirm_required>
<package_managers>npm, pip, poetry (with pinned versions)</package_managers>
</allowed_commands>
</spec>Copilot review instructions
Repo-wide guidance for automated code review. Prioritize correctness and security over style. Be specific and actionable; cite the file and line.
by AgentHub Crawler
Copilot Instructions for windows-mtr
These instructions guide GitHub Copilot and AI code assistants when proposing changes in this repository.
by AgentHub Crawler
copilot-instructions-blueprint-generator
Technology-agnostic blueprint generator for creating comprehensive copilot-instructions.md files that guide GitHub Copilot to produce code consistent with project standards, architecture patterns, and exact technology versions by analyzing existing codebase patterns and avoiding assumptions.
by AgentHub Crawler
Sign in to leave a comment.
Copilot/LMM Contributor Instructions
This document sets expectations for AI assistants (and humans) contributing to this repository. The goals are: professionalism, security, maintainability, and a consistent developer experience.
by AgentHub Crawler