Skip to content

handoff-2026-08-02.md

Latest commit

 

History

History
176 lines (137 loc) · 5.89 KB

handoff-2026-08-02.md

File metadata and controls

176 lines (137 loc) · 5.89 KB
Raw

MisakaNet current handoff

Repo: C:\Users\hp\MisakaNet Maintainer: Ikalus1988 Date: 2026-08-02

Current state

  • v2.14.0 released: PyPI, GitHub Release, GHCR container.
  • GHCR images:
    • ghcr.io/ikalus1988/misakanet:2.14.0
    • ghcr.io/ikalus1988/misakanet:latest
  • PR Genius merged in #722:
    • advisory only, not merge blocker
    • pinned to v1.0.0
    • permissions: contents: read
    • skips draft/docs-only PRs
    • no PR comments, only CI logs
  • Welcome workflows cleaned:
    • pure ASCII
    • DCO-first
    • Search/Register/CONTRIBUTING links
    • pr-welcome.yml permissions: issues: write + pull-requests: read

Recently completed

  • SEO landing pages merged:
    • #736 MCP setup SEO
    • #738 Windows encoding SEO
  • Closed-DCO absorption batch completed:
    • #748 shell helper (#741)
    • #747 CI hygiene pycache reject (#746)
    • #750 English translations (#743)
    • #749 benchmark task catalog (#742)
    • #754 query expansion (#745)
    • #752 intake digest CLI (#744)
    • #740 runtime smoke report (#683)
  • Issues #741-#746 and #683 are closed.
  • Verification:
    • 433 tests passed
    • 1 pre-existing failure: test_ci_self_heal
    • intake_digest.py --help works
    • misaka_run.py --help works
    • misaka-search.sh is a bash script and should not be run with Python

Current open issue structure

Active / next:

  • #756 PR Genius observation over 10 PRs
  • #757 Runtime smoke matrix: Cursor / Claude Code / shell / GHCR / Glama path
  • #758 GHCR container quickstart docs
  • #759 Landing page links from high-signal issues
  • #760 Glama first MCP tool call conversion, blocked by #764 evidence
  • #761 Public MCP smoke report
  • #764 Verify Glama Tool Calls analytics counting

Existing later/backlog:

  • #682 benchmark
  • #552 Feishu duplicate cleanup
  • #513 danmaku experiment
  • #429 homepage blind test
  • #319 MCP protocol extension
  • #646 onboarding
  • #309 translation

Important metrics / wording

Glama analytics for 2026-07-07 to 2026-08-02:

  • Impressions: 80
  • Clicks: 8
  • CTR: 10%
  • Profile Views: 1,433
  • Tool Calls: 0

Interpretation:

  • Tool Calls = 0 does not necessarily mean nobody uses MisakaNet.
  • Likely Glama counts only Glama-routed / hosted gateway calls, not local stdio calls.
  • #764 was created to verify the analytics counting boundary before over-investing in #760.

Use wording:

  • Say: 0 Glama-routed tool calls
  • Do not say: 0 usage

DeepSeekHarness opportunity

Tianyi Cui posted:

If you are an Agent Harness related OSS project developer and want to join DeepSeekHarness beta, reply/DM with GitHub ID and representative OSS work.

MisakaNet positioning:

  • MisakaNet is not a full agent harness.
  • Best wording: failure-memory and recovery layer for coding-agent harnesses and MCP clients.

Suggested pitch:

  • GitHub ID: Ikalus1988
  • Representative work: MisakaNet
  • Repo: https://github.com/Ikalus1988/MisakaNet
  • Optional MCP listing: https://glama.ai/mcp/servers/Ikalus1988/MisakaNet

MisakaNet can provide DeepSeekHarness:

  1. DeepSeekHarness x MisakaNet integration report
  2. MCP tool-calling smoke results
  3. failure recovery smoke matrix
  4. 10-task coding-agent failure recovery benchmark catalog
  5. context/tool-result injection feedback
  6. optional public integration note

Suggested short reply:

想报名 DeepSeekHarness 内测。

GitHub ID: Ikalus1988
代表作: MisakaNet
https://github.com/Ikalus1988/MisakaNet

MisakaNet 是面向 coding agents / MCP clients / agent harness 的 failure-memory & recovery layer，提供 MCP tools、CLI、GHCR、Cursor/Claude Code/shell runtime entry，用于让 agent 在遇到工程失败时检索可复用 recovery lessons。

很希望参与内测，并反馈 MCP tool calling、failure recovery、context reuse、runtime smoke 等真实问题。

Maintainer policy

  • No DCO, no review/merge.
  • Closed DCO-failed PRs can inspire clean-room issues, but code/text should not be copied.
  • Keep PRs small and scoped.
  • PR Genius is advisory triage signal only.
  • DCO and audit/shape/security remain hard blockers.
  • Avoid overclaiming:
    • no verified users
    • no fake benchmark scores
    • no MisakaNet is an agent harness

Recommended next steps

  1. Reply/DM for DeepSeekHarness beta.
  2. Do #764 first: verify what Glama Tool Calls actually counts.
  3. Do #761/#757 smoke docs after #764 evidence.
  4. Keep #760 blocked until Glama counting boundary is known.
  5. Observe PR Genius across 5-10 PRs before changing workflow again.

Continuation log — 2026-08-02 17:50 CST

  • Local docs drafted for #764:
    • docs/integrations/glama-analytics.md
    • Evidence: Glama API listing https://glama.ai/api/mcp/v1/servers/Ikalus1988/MisakaNet returns tools: [].
    • Current conclusion: say 0 Glama-routed tool calls; do not say 0 usage.
  • Local docs drafted for #761:
    • docs/integrations/mcp-smoke-report.md
    • Verified stdio MCP calls: initialize, tools/list, misakanet_search, misakanet_usage_status.
  • Local implementation cleanup:
    • scripts/mcp_server.py now reports serverInfo.version from installed metadata or pyproject.toml instead of stale 2.12.0.
    • tests/test_mcp_server.py now checks MCP serverInfo.version against pyproject.toml.
  • Integration index linked:
    • docs/integrations/README.md links the MCP smoke report and Glama analytics boundary doc.
  • Plugin/handoff update:
    • misakanet-growth skill source and installed cache now point to this handoff and include the Glama wording rule.
  • Validation:
    • python -m pytest tests/test_mcp_server.py -q -> 9 passed.
    • python tests/test_mcp_server.py -> 31 passed, 0 failed.
    • python C:\Users\hp\plugins\misakanet-growth\scripts\misakanet_growth.py smoke -> PASS.
    • plugin manifest validation -> PASS.
    • skill validation -> PASS.
    • site health -> watchlist clear.

Remaining local checkout note

  • misakanet/profile.json was already dirty from local usage counter/activity changes and is not part of the docs/MCP evidence work.