Add comprehensive onboarding guide for AI coding agents

[Copilot]

Expanded .github/copilot-instructions.md from minimal placeholder to comprehensive 408-line onboarding guide for AI coding agents working in the CCF repository.

Sections Added

• Repository structure - Key directories, languages, and what CCF is
• Build system - CMake + Ninja commands, options (BUILD_TESTS, SAN, CLANG_TIDY), 32GB RAM requirement
• Testing - Doctest (C++) and pytest (E2E) infrastructure, importance of ./tests.sh wrapper for venv setup
• Code formatting - scripts/ci-checks.sh -f enforcement for clang-format, ruff, prettier
• Conventions - C++/Python naming, logging macros (LOG_*_FMT), error handling patterns
• Development workflows - DevContainer setup, CI pipeline stages, documentation building
• Common issues - Memory exhaustion workarounds, port conflicts, formatting failures
• Security - MSRC reporting, no secrets in code, crypto library usage
• CI/CD - PR requirements, extended test labels (run-long-test, bench-ab)
• Quick reference - Command cheatsheet, critical files, help resources

Key Requirements Documented

# Build (32GB+ RAM required)
cmake -GNinja .. && ninja

# Test (must use wrapper for Python venv)
./tests.sh

# Format before commit (CI blocks unformatted code)
./scripts/ci-checks.sh -f

Shell Script Requirement

Codified existing rule: scripts using pipes must set set -o pipefail to catch mid-pipeline failures.

---

Note: Commit includes minor C++ formatting fixes (whitespace before {};) auto-applied by ci-checks.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[maxtropets]

How is this of any use for a robot actually? Can we have a short human-readable bullet points file, which can be easily review and extended instead?

[achamayou]

@maxtropets I suggest skimming https://arxiv.org/abs/2602.11988

[maxtropets]

@maxtropets I suggest skimming https://arxiv.org/abs/2602.11988

Well, the abstract says the same?..

Across multiple coding agents and LLMs, we
find that context files tend to reduce task success
rates compared to providing no repository
context, while also increasing inference cost by
over 20%. Behaviorally, both LLM-generated
and developer-provided context files encourage
broader exploration (e.g., more thorough testing
and file traversal), and coding agents tend to
respect their instructions. Ultimately, we conclude
that unnecessary requirements from context files
make tasks harder, and human-written context
files should describe only minimal requirements

[eddyashton]

@maxtropets I suggest skimming https://arxiv.org/abs/2602.11988

Interesting source. I think the mention of increased project-specific tool use and testing is interesting, and something we'd really benefit from. A drop in success rate is annoying, but hard to measure. The citation on the benefits of a security prompt looks valuable too.

On balance, I think these auto-generated instructions are too long (bloating the context window), unfocussed, and occasionally outright misleading (ie - mention pytest, say all PRs must come from a fork). I think we would benefit from some more repo-specific guidance (eg - see #7676, where it first tries a standalone pytest and example script, both untested, then rolls into an existing e2e test but doesn't run it). I'll hand-write something more focussed.