fix(pipeline): stop incremental context runs gutting updated sources

fix(frontend): number citations whose labels contain balanced brackets

fix(frontend): source info popover tracks scroll

fix(activity): repeated Gemini search queries start a fresh cycle

fix(graphrag): query-relevant community reports, rerank floor parity

fix(retrieval): budget skip, standard RRF, fail-fast collection load

fix(config): restore default dotenv precedence (shell wins over .env)

fix(config): honor AI_TUTOR_KB_DIR for every KB path

fix(citations): keep labels of links with balanced brackets

perf(citations): cache the KB manifest indexes

fix(api): close request-size gaps in body and history caps

fix(api): emit stream error frame after cleanup parts

chore(frontend): TypeScript 6.0.3 + lucide-react 1.23.0

chore(deps): refresh backend lock + frontend deps; Node 22 builder

fix(citations): handle malformed links; count only consulted docs in activity (#12)

Public docs-only bundle: allowlist filtering, leak audit, kb.tar.gz cold start (#11)

wip: eval harness + public docs bundle; relocate eval docs into evals/ (#10)

chore: add LICENSE, security policy, and issue/PR templates (#9)

feat(registry): rename agentic_ai_engineering label to "Agent Engineering: Building Multi-Agent Systems"

fix(citations): drop code-block URLs from chips; default URL reading off

feat(kb): register guardrails and serving topic pages in wiki scaffolder

Add 10-Hour LLM Primer course as a new KB source

feat(prompts): add course content reproduction policy and related tests

chore: remove outdated context engineering documentation files

docs(evals): clarify F29's full_history arm is a document-in-prompt, not the conversational policy

Merge experiment #3: SLM knowledge compaction (Ollama small models, 32k window)

docs: add evals_contributing.md — teammate experiment playbook

docs(readme): link the context-engineering experiments Space + update slide deck

docs(evals): F29 — context-compaction maps a boundary where full_history stops winning

Merge experiment #5: context-compaction study (keep-all vs compaction vs retrieve, Gemini 2.5 Flash)

build(graphrag): make graphrag an eval-only optional extra (keep prod image lean)

docs(evals): promote GraphRAG result to F28 in the central findings log

Merge experiment #2: GraphRAG vs classical RAG (opt-in retriever + scoped comparison)

feat(evals): single-turn latency finding (F27) + de-confound latency tooling

chore: gitignore deepseek_eval_results.md (local results summary)

evals(slm): point the writeup at the shared HF dataset for data + results

evals(slm): Axis-A results (3 SLMs x 8 memory presets) + F25, dual-experiment writeup

evals(compaction-study): skip non-run dirs in report (fix regen crash)

evals(slm): longer turn timeout for slow SLM map-reduce + model-aware report

evals(slm): Axis-A memory-compaction arm on local SLMs

evals: SLM knowledge-compaction findings (F24) + writeup

evals(knowledge-compaction): harden judge against truncated JSON + re-judge tool

evals: 3-model SLM runner and cross-model comparison report

evals(knowledge-compaction): selectable strategies + local SLM (Ollama) arm

compaction study: standalone 2.5-Flash results writeup + report glob fix

evals(compaction-study): report Family A presets vs Family B retrieval

memory: add hierarchical and delta summarization presets (Axis A)

feat: enhance agent instructions with support for DeepSeek and OpenRouter in model configuration

feat: update environment variables for OpenRouter and DeepSeek integration

feat: add findings on DeepSeek-V4-Flash performance and memory strategies