[Changyong Gong (chagong)]

Summary

• add a patch-free pull request file manifest tool that projects REST metadata before the model-result limit
• require an explicit component/evidence plan before reading patches or source
• prefer pinned source reads and bounded five-file patch groups instead of one file per model turn
• preserve anonymous reads for public repositories without an IssueLens App installation

Why

The #57 team-memory run forced 23 separate list_pull_request_files(per_page=1) model turns. Repeatedly replaying the growing conversation drove the 125k TPM deployment into likely throttling. For the same PR, the new manifest returns all 23 file entries in one approximately 13 KB model payload rather than 23 patch-bearing discovery turns.

This follows GitHub Copilot code review's published large-PR pattern: make an explicit plan, explore relevant repository files with tools, and capture findings while reading rather than deferring all reasoning until the end.

Validation

• 244 MCP tests passed
• 404 runnable application tests passed
• Python compilation passed
• git diff --check passed
• live GitHub verification confirmed PR Add atomic wiki page patching #57's 23-file REST response projects from about 79 KB to about 13 KB without patches

[Copilot]

🟡 Changes recommended

Manifest pagination can advertise an invalid continuation page, and the new tool lacks telemetry classification.

2 open issues

• Fix pagination at the maximum supported page boundary · New
• Register pull request manifest calls in telemetry READ_TOOLS · New

What changed in this PR

Optimizes IssueLens large-PR analysis by separating compact file discovery from targeted evidence reads, reducing repeated patch content in model conversations.

Changes:

• Adds a patch-free PR manifest tool with bounded responses and anonymous public-repository fallback.
• Updates agents to plan evidence reads before inspecting pinned source or bounded patch groups.
• Extends tool registration, tests, and documentation.

File	Description
tests/​test_team_memory_push.py	Checks updated batch-analysis instructions.
tests/​test_team_memory_instructions.py	Updates the maintenance tool allowlist expectation.
tests/​test_team_memory_coordinator.py	Checks manifest-first coordinator requests.
tests/​test_mcp_tool_documentation.py	Adds manifest permission expectations.
tests/​test_agent_prompts.py	Checks evidence-planning guidance.
skills/​change-analysis/​SKILL.md	Defines manifest-first analysis and bounded evidence reads.
main.py	Allows the maintenance agent to use manifests.
github_app_mcp/​tests/​test_stdio.py	Updates expected tool discovery.
github_app_mcp/​tests/​test_server.py	Tests manifest schema and argument forwarding.
github_app_mcp/​tests/​test_github.py	Tests projection, pagination, validation, and anonymous fallback.
github_app_mcp/​src/​issuelens_github_mcp/​server.py	Exposes the manifest tool.
github_app_mcp/​src/​issuelens_github_mcp/​github.py	Implements manifest retrieval and metadata projection.
github_app_mcp/​README.md	Documents manifest usage and limits.
docs/​guide.md	Explains planned evidence retrieval.
agents/​triage.md	Adopts manifest-first investigation.
agents/​team-memory.md	Requires evidence planning for maintenance.
agents/​plan.md	Updates planning evidence guidance.
agents/​issuelens.md	Describes the revised analysis workflow.
.github/​copilot-instructions.md	Updates architecture documentation.
.github/​actions/​issuelens/​issuelens_action.py	Generates manifest-first batch requests.

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

Static review identified no blocking issues, and focused regression coverage addresses the new tool and its integrations.

No open issues 🎉

2 resolved since last review

• Register pull request manifest calls in telemetry READ_TOOLS
• Fix pagination at the maximum supported page boundary

_{Review effort: Balanced}