New docs/tool-result-payloads.md, a decision record in the same style as memory-api-improvements.md. It covers:
why every tool result repeats its payload as JSON text (claude.ai ignores structuredContent), and why client detection and structured-only responses were rejected
why memory_list / memory_search return a ~200-char excerpt with contentLength and truncated, the full: true option, and why the field isn't named snippet
why excerpts weight rare query terms instead of using ts_headline, including the tested ts_headline output and what would reopen the question
invariants to preserve
README: linked from Design notes, plus two troubleshooting entries: claude.ai getting only "N result(s)", and list/search showing an excerpt instead of the full text.
docs/memory-api-improvements.md: marked the "same 9 fields as list/search" claim as superseded.
tools.ts: corrected the same claim in a memory.get code comment. This is a comment-only change, so behaviour is unchanged.
Docs for #27, #28 and #29.
- **New `docs/tool-result-payloads.md`**, a decision record in the same style as `memory-api-improvements.md`. It covers:
- why every tool result repeats its payload as JSON text (claude.ai ignores `structuredContent`), and why client detection and structured-only responses were rejected
- why `memory_list` / `memory_search` return a ~200-char `excerpt` with `contentLength` and `truncated`, the `full: true` option, and why the field isn't named `snippet`
- why excerpts weight rare query terms instead of using `ts_headline`, including the tested `ts_headline` output and what would reopen the question
- invariants to preserve
- **README**: linked from Design notes, plus two troubleshooting entries: claude.ai getting only "N result(s)", and list/search showing an excerpt instead of the full text.
- **`docs/memory-api-improvements.md`**: marked the "same 9 fields as list/search" claim as superseded.
- **`tools.ts`**: corrected the same claim in a `memory.get` code comment. This is a comment-only change, so behaviour is unchanged.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Adds docs/tool-result-payloads.md covering PRs #27-#29: why every tool
result repeats its payload as JSON text (claude.ai ignores
structuredContent; client detection and structured-only were rejected),
why memory_list / memory_search return excerpts (named to avoid the
snippet.* tools), and why excerpts weight rare query terms rather than
using ts_headline (tested; it picked the same wrong passage).
Links it from the README design notes, adds two troubleshooting entries,
and marks the "same 9 fields as list/search" claims in the earlier
decision record and in memory.get's comment as superseded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jknapp
merged commit 03eb0ab85e into main2026-11-14 19:25:20 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Docs for #27, #28 and #29.
docs/tool-result-payloads.md, a decision record in the same style asmemory-api-improvements.md. It covers:structuredContent), and why client detection and structured-only responses were rejectedmemory_list/memory_searchreturn a ~200-charexcerptwithcontentLengthandtruncated, thefull: trueoption, and why the field isn't namedsnippetts_headline, including the testedts_headlineoutput and what would reopen the questiondocs/memory-api-improvements.md: marked the "same 9 fields as list/search" claim as superseded.tools.ts: corrected the same claim in amemory.getcode comment. This is a comment-only change, so behaviour is unchanged.🤖 Generated with Claude Code