Skip to content

feat(v0.5.1): universal pre-enrichment, compact context format, prose…#2

Merged
Divish1032 merged 5 commits intomainfrom
v0.5.1
Mar 4, 2026
Merged

feat(v0.5.1): universal pre-enrichment, compact context format, prose…#2
Divish1032 merged 5 commits intomainfrom
v0.5.1

Conversation

@Divish1032
Copy link
Collaborator

… briefings

  • bulkIngestToBrain: remove 100-node cap, ingest ALL non-structural nodes, increase concurrency 4→8 workers (0.8B handles more parallel requests)
  • add get_context format="compact" parameter: serializeCompact() converts directional context to natural-language prose ~400-600 tokens (80% reduction vs JSON default of 2000-3800 tokens); new internal/mcp/digest.go
  • register format param in get_context tool definition (server.go)
  • update CLAUDE.md with new package table and session_init startup ritual

Divish1032 and others added 5 commits March 3, 2026 17:49
… briefings

- bulkIngestToBrain: remove 100-node cap, ingest ALL non-structural nodes,
  increase concurrency 4→8 workers (0.8B handles more parallel requests)
- add get_context format="compact" parameter: serializeCompact() converts
  directional context to natural-language prose ~400-600 tokens (80% reduction
  vs JSON default of 2000-3800 tokens); new internal/mcp/digest.go
- register format param in get_context tool definition (server.go)
- update CLAUDE.md with new package table and session_init startup ritual

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Covers all 3 legs (synapses core, intelligence, scout) with pass/fail
checks, token savings methodology, scoring rubric, and live results
from the v0.5.1 run. Re-run this plan for every major version to
track regression vs improvement.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
19 improvements across P0-P3 covering: CPU model selection, get_impact
overflow, hot constitution, ADRs, domain personas, retrieval hooks,
progressive context loading, and federated cold memory.

Maps each gap to the 3-tier Codified Context framework (arxiv 2602.20478).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…context loading

Batch 1 (quick wins):
- BUG-S02: get_impact initialises Tiers to [] instead of null
- BUG-S04: get_context adds brain hint when brain is not configured
- UX-S04: get_file_context groups entities by file when multiple files match
- UX-S05: get_working_state scans root dir for .json/.yaml/.toml/Makefile/Dockerfile
- get_impact node cap + hot-reload synapses.json + call-chain cross-binary explanation (v0.5.2)
- FTS prefix matching + find_entity test-file ranking (v0.5.2)

Batch 3 (medium features):
- P2-D: detail_level param on get_context(format=compact): summary/neighbors/full
- Constitution: synapses.json constitution.principles → injected in session_init + get_context compact

Batch 4 (ADRs):
- upsert_adr + get_adrs MCP tools backed by brain /v1/adr HTTP routes
- ADR lines injected into get_context compact when brain is configured
- brain.Client: UpsertADR(), GetADR(), GetADRs() HTTP methods
- config: ConstitutionConfig struct + applyDefaults injection flags

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@Divish1032 Divish1032 merged commit ea20da2 into main Mar 4, 2026
2 of 4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant