[GH-ISSUE #8293] refactor(evidence): unify declared Evidence inputs across workflows #6529

New issue

Closed

opened 2026-11-01 21:11:58 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-11-01 21:11:58 +00:00

Owner

Originally created by @gensokyo-dev on GitHub (Oct 31, 2026).
Original GitHub issue: https://github.com/morluto/rea/issues/8293

Before submitting

I searched open and closed issues and pull requests for related proposals.
I understand the contribution scope must be reviewed before integration.
This request contains no credentials, private samples, or sensitive target data.

Area

Evidence, artifacts, or reconstruction; CLI and MCP integration.

Current limitation

Evidence reuse remains opt-in per tool. evidenceInputSchema and exact-reference resolution exist, but schemas, source resolution, semantic admission, and source retention are wired separately. Some consumers accept retained IDs while others require complete records. #8281 reproduces a concrete two-dossier build_call_path failure at the default MCP input allowance. Recent improvements in #8214 and #7678 should be preserved and reused.

Source inspection at main 23de10bb2 also shows separate native, JavaScript, managed, WASM, and artifact input paths. This is source-derived scope; broader real-provider and performance coverage has not yet been executed for this proposal.

Desired outcome and success criteria

Use one declared Evidence-input contract and shared application resolution path across Evidence-consuming tools, while keeping domain payload decoders and compatibility/coverage rules with their owners. Existing inline inputs remain supported; exact retained references avoid resending complete records. Shared declarations should drive advertised schemas and actual input handling so new consumers cannot silently omit reference support.

Preserve exact IDs, caller selections, provenance, partial/unknown coverage, source closure, existing capacity/cancellation controls and complete Evidence access. Missing IDs must fail explicitly without rerunning a provider. CLI/MCP share the same implementation where both entrypoints already exist; MCP-only operations need no new CLI commands. Whole-bundle and raw-observation APIs retain their distinct semantics.

Expected interaction with targets

Read existing files or captured data where the existing contract already permits it. No new providers, execution, injection, mutation, or protection-recovery capability is proposed. Do not expand file-reading authority merely to unify input shapes.

Verification available so far

Related real Ghidra reproduction: #8281. Planned checks: public producer-summary-to-consumer reference journeys, inline/reference equivalence, unavailable/wrong-kind references, independent inline admission, identity/closure/coverage and cancellation regressions, CLI/MCP and converted JSON Schema checks. Exercise affected domains with real public applications and local Codex CLI gpt-6.1-terra / high, with bounded CPU/process-family memory and honest usage/time/CPU/memory accounting.

#8281, #8214, #7678, #8172, #8237. This tracks the common owning mechanism and consumer migration; it does not replace the related reports or claim they share one runtime cause.

Originally created by @gensokyo-dev on GitHub (Oct 31, 2026). Original GitHub issue: https://github.com/morluto/rea/issues/8293 ### Before submitting - [x] I searched open and closed issues and pull requests for related proposals. - [x] I understand the contribution scope must be reviewed before integration. - [x] This request contains no credentials, private samples, or sensitive target data. ### Area Evidence, artifacts, or reconstruction; CLI and MCP integration. ### Current limitation Evidence reuse remains opt-in per tool. `evidenceInputSchema` and exact-reference resolution exist, but schemas, source resolution, semantic admission, and source retention are wired separately. Some consumers accept retained IDs while others require complete records. #8281 reproduces a concrete two-dossier `build_call_path` failure at the default MCP input allowance. Recent improvements in #8214 and #7678 should be preserved and reused. Source inspection at main `23de10bb2` also shows separate native, JavaScript, managed, WASM, and artifact input paths. This is source-derived scope; broader real-provider and performance coverage has not yet been executed for this proposal. ### Desired outcome and success criteria Use one declared Evidence-input contract and shared application resolution path across Evidence-consuming tools, while keeping domain payload decoders and compatibility/coverage rules with their owners. Existing inline inputs remain supported; exact retained references avoid resending complete records. Shared declarations should drive advertised schemas and actual input handling so new consumers cannot silently omit reference support. Preserve exact IDs, caller selections, provenance, partial/unknown coverage, source closure, existing capacity/cancellation controls and complete Evidence access. Missing IDs must fail explicitly without rerunning a provider. CLI/MCP share the same implementation where both entrypoints already exist; MCP-only operations need no new CLI commands. Whole-bundle and raw-observation APIs retain their distinct semantics. ### Expected interaction with targets Read existing files or captured data where the existing contract already permits it. No new providers, execution, injection, mutation, or protection-recovery capability is proposed. Do not expand file-reading authority merely to unify input shapes. ### Verification available so far Related real Ghidra reproduction: #8281. Planned checks: public producer-summary-to-consumer reference journeys, inline/reference equivalence, unavailable/wrong-kind references, independent inline admission, identity/closure/coverage and cancellation regressions, CLI/MCP and converted JSON Schema checks. Exercise affected domains with real public applications and local Codex CLI `gpt-6.1-terra` / high, with bounded CPU/process-family memory and honest usage/time/CPU/memory accounting. ### Related work #8281, #8214, #7678, #8172, #8237. This tracks the common owning mechanism and consumer migration; it does not replace the related reports or claim they share one runtime cause.

bsakaguchi closed this issue

2026-11-01 21:11:58 +00:00

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

morluto/rea#6529

No description provided.

Rows
Columns