Compaction: preserve opaque identifiers in summaries by rodrigouroz · Pull Request #25553 · openclaw/openclaw

rodrigouroz · 2026-02-24T15:26:18Z

Summary

Describe the problem and fix in 2–5 bullets:

Problem: compaction summaries could drift or reconstruct opaque identifiers (IDs, URLs, hashes) instead of preserving exact values.
Why it matters: post-compaction context can become unreliable when exact identifiers change.
What changed: compaction summarization now always prepends identifier-preservation instructions; added focused tests; added one changelog entry.
What did NOT change (scope boundary): no safeguard-mode quality guard/retry logic, no post-compaction memory sync behavior.

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

Closes n/a (final close tracked in Compaction Runner: wire post-compaction memory sync #25561)
Related [Bug]: Compaction summarization does not preserve exact identifiers #19207, (fix): Compaction: preserve recent context and sync session memory post-compact #20038

User-visible / Behavior Changes

Compaction summaries are instructed to preserve opaque identifiers verbatim by default.

Security Impact (required)

New permissions/capabilities? (No)
Secrets/tokens handling changed? (No)
New/changed network calls? (No)
Command/tool execution surface changed? (No)
Data access scope changed? (No)
If any Yes, explain risk + mitigation:

Repro + Verification

Environment

OS: macOS
Runtime/container: Node 22 + pnpm
Model/provider: n/a (unit tests)
Integration/channel (if any): n/a
Relevant config (redacted): default

Steps

pnpm vitest run src/agents/compaction.identifier-preservation.test.ts
pnpm tsgo
Verify changed files are limited to compaction prompt path + tests/changelog.

Expected

Identifier-preservation tests pass and typecheck remains clean.

Actual

Local tests passed; pnpm tsgo passed.

Evidence

Attach at least one:

Failing test/log before + passing after
Trace/log snippets
Screenshot/recording
Perf numbers (if relevant)

Human Verification (required)

What you personally verified (not just CI), and how:

Verified scenarios: instruction text is injected with and without custom instructions; split/merge summarization keeps the same instruction.
Edge cases checked: custom instructions still append under "Additional focus".
What you did not verify: live provider behavior in long-running sessions.

Compatibility / Migration

Backward compatible? (Yes)
Config/env changes? (No)
Migration needed? (No)
If yes, exact upgrade steps:

Failure Recovery (if this breaks)

How to disable/revert this change quickly: revert this PR.
Files/config to restore: src/agents/compaction.ts
Known bad symptoms reviewers should watch for: none expected; only prompt guidance changed.

Risks and Mitigations

List only real risks for this PR. Add/remove entries as needed. If none, write None.

Risk: stronger prompt constraints could alter summary style.
- Mitigation: scope is limited to identifier retention and covered by focused tests.

Greptile Summary

This PR hardens compaction summarization to preserve opaque identifiers (UUIDs, IDs, tokens, hostnames, IPs, ports, URLs, file names) by prepending identifier-preservation instructions to all summarization prompts.

Key changes:

Added buildCompactionSummarizationInstructions() helper that always prepends identifier-preservation guidance before any custom instructions
Modified summarizeChunks() to use buildCompactionSummarizationInstructions() instead of passing raw customInstructions
Fixed merge-path instruction handling in summarizeInStages() to avoid duplicating "Additional focus:" headers
Added comprehensive test coverage for all instruction injection paths
Unrelated fix: added explicit type parameter <string> to Set in ui/src/ui/external-link.ts:7

Confidence Score: 5/5

This PR is safe to merge with minimal risk
The changes are well-scoped to prompt instruction assembly with comprehensive test coverage. The implementation correctly prepends identifier-preservation guidance to all summarization paths (chunk, split, and merge). The merge-path fix properly avoids duplicating headers. The unrelated TypeScript typing improvement is a harmless clarification. Tests verify all edge cases including split/merge paths and custom instruction handling.
No files require special attention

_{Last reviewed commit: c4bf3ff}

greptile-apps

_{3 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-02-24T15:29:49Z

CHANGELOG.md

 - Ollama/Qwen: handle Qwen 3 reasoning field format in Ollama responses. (#18631) Thanks @mr-sk.
 - OpenAI/Transcripts: always drop orphaned reasoning blocks from transcript repair. (#18632) Thanks @TySabs.
 - Fix types in all tests. Typecheck the whole repository.
+- Agents/Compaction: harden summarization prompts to preserve opaque identifiers verbatim (UUIDs, IDs, tokens, host/IP/port, URLs), reducing post-compaction identifier drift and hallucinated identifier reconstruction.


Changelog entry in released version section

This new entry is placed under the 2026.2.17 section (an already-released version) rather than Unreleased or 2026.2.23 (Unreleased) at the top. Is this intentional for the PR stack ordering, or should it be moved to the unreleased section?

Prompt To Fix With AI

This is a comment left during a code review. Path: CHANGELOG.md Line: 624 Comment: **Changelog entry in released version section** This new entry is placed under the `2026.2.17` section (an already-released version) rather than `Unreleased` or `2026.2.23 (Unreleased)` at the top. Is this intentional for the PR stack ordering, or should it be moved to the unreleased section? How can I resolve this? If you propose a fix, please make it concise.

Good catch. I moved this entry from the released 2026.2.17 block to 2026.2.24 (Unreleased) under ### Fixes

rodrigouroz · 2026-02-24T16:31:38Z

@greptileai can you re verify and update the PR description?

greptile-apps

_{4 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-02-24T16:35:17Z

Additional Comments (1)

src/agents/compaction.ts
Duplicate "Additional focus:" header in merge path

When params.customInstructions is set and the split+merge path fires, mergeInstructions is built with an "Additional focus:\n" wrapper (line 339), then passed as customInstructions to summarizeWithFallback → summarizeChunks → buildCompactionSummarizationInstructions, which adds a second "Additional focus:\n" wrapper (line 27). The resulting prompt looks like:

Preserve all opaque identifiers exactly as written …

Additional focus:
Merge these partial summaries into a single cohesive summary. …

Additional focus:
<user's custom instructions>

This won't break anything at runtime, but the redundant header is noisy. Consider either:

Having buildCompactionSummarizationInstructions handle the merge case too (accepting a structured options object), or
Wrapping the merge instructions through buildCompactionSummarizationInstructions explicitly and appending merge-specific text separately.

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/agents/compaction.ts
Line: 338-340

Comment:
**Duplicate "Additional focus:" header in merge path**

When `params.customInstructions` is set and the split+merge path fires, `mergeInstructions` is built with an `"Additional focus:\n"` wrapper (line 339), then passed as `customInstructions` to `summarizeWithFallback` → `summarizeChunks` → `buildCompactionSummarizationInstructions`, which adds a second `"Additional focus:\n"` wrapper (line 27). The resulting prompt looks like:

```
Preserve all opaque identifiers exactly as written …

Additional focus:
Merge these partial summaries into a single cohesive summary. …

Additional focus:
<user's custom instructions>
```

This won't break anything at runtime, but the redundant header is noisy. Consider either:
- Having `buildCompactionSummarizationInstructions` handle the merge case too (accepting a structured options object), or
- Wrapping the merge instructions through `buildCompactionSummarizationInstructions` explicitly and appending merge-specific text separately.

How can I resolve this? If you propose a fix, please make it concise.

rodrigouroz · 2026-02-24T16:40:33Z

Thanks @greptileai, I added a regression test with your suggestion. Please edit the PR description after you have verified again.

…penclaw#25553) Hardens summarization prompts to retain UUIDs, IDs, tokens, host/IP/port, URLs, and file names verbatim, reducing post-compaction identifier drift and hallucinated identifier reconstruction. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Takhoffman · 2026-02-27T14:14:15Z

PR #25553 - Compaction: preserve opaque identifiers in summaries (#25553)

Merged via squash.

Merge commit: 0fe6cf0
Verified: pnpm install --frozen-lockfile, pnpm build, pnpm check, pnpm test:macmini
Changes made:
CHANGELOG.md
docs/concepts/compaction.md
docs/gateway/configuration-reference.md
src/agents/compaction.identifier-policy.test.ts
src/agents/compaction.identifier-preservation.test.ts
src/agents/compaction.ts
src/agents/pi-embedded-runner/extensions.ts
src/agents/pi-extensions/compaction-safeguard-runtime.ts
src/agents/pi-extensions/compaction-safeguard.ts
src/config/config.compaction-settings.test.ts
src/config/schema.help.quality.test.ts
src/config/schema.help.ts
src/config/schema.labels.ts
src/config/types.agent-defaults.ts
src/config/zod-schema.agent-defaults.ts
Why these changes were made:
Make compaction identifier-preservation behavior configurable while keeping strict identifier retention as the default.
Changelog: CHANGELOG.md updated=true required=true opt_out=false

Thanks @rodrigouroz!

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com> (cherry picked from commit 3b0f73e)

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com> (cherry picked from commit 3b0f73e)

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com> (cherry picked from commit 3b0f73e)

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

@rodrigouroz

… thanks @rodrigouroz Verified: - pnpm install --frozen-lockfile - pnpm build - pnpm check - pnpm test:macmini Co-authored-by: rodrigouroz <384037+rodrigouroz@users.noreply.github.com> Co-authored-by: Tak Hoffman <781889+Takhoffman@users.noreply.github.com>

openclaw-barnacle bot added agents Agent runtime and tooling size: M experienced-contributor labels Feb 24, 2026

greptile-apps bot reviewed Feb 24, 2026

View reviewed changes

rodrigouroz force-pushed the codex/pr20038-01 branch from e279efb to 2e50696 Compare February 24, 2026 15:36

openclaw-barnacle bot added size: S and removed size: M labels Feb 24, 2026

rodrigouroz mentioned this pull request Feb 24, 2026

(fix): Compaction: preserve recent context and sync session memory post-compact #20038

Closed

18 tasks

openclaw-barnacle bot added the app: web-ui App: web-ui label Feb 24, 2026

rodrigouroz force-pushed the codex/pr20038-01 branch from 7cc85d1 to 15ef409 Compare February 24, 2026 16:27

Compaction: preserve opaque identifiers in summaries

140b9be

rodrigouroz force-pushed the codex/pr20038-01 branch from 42cc7fb to 140b9be Compare February 24, 2026 16:33

greptile-apps bot reviewed Feb 24, 2026

View reviewed changes

Compaction: avoid duplicate merge focus header

c4bf3ff

Merge branch 'main' into codex/pr20038-01

10588a1

openclaw-barnacle bot removed the app: web-ui App: web-ui label Feb 24, 2026

rodrigouroz and others added 2 commits February 24, 2026 17:58

Merge branch 'main' into codex/pr20038-01

442169b

Config: add compaction identifier policy with strict default

0508650

openclaw-barnacle bot added docs Improvements or additions to documentation gateway Gateway runtime size: M and removed size: S labels Feb 27, 2026

Takhoffman merged commit 0fe6cf0 into openclaw:main Feb 27, 2026
29 of 30 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Compaction: preserve opaque identifiers in summaries#25553

Compaction: preserve opaque identifiers in summaries#25553
Takhoffman merged 5 commits intoopenclaw:mainfrom
rodrigouroz:codex/pr20038-01

rodrigouroz commented Feb 24, 2026 •

edited by greptile-apps bot

Loading

Uh oh!

greptile-apps bot left a comment

Uh oh!

greptile-apps bot Feb 24, 2026

Uh oh!

rodrigouroz Feb 24, 2026

Uh oh!

rodrigouroz commented Feb 24, 2026

Uh oh!

greptile-apps bot left a comment

Uh oh!

greptile-apps bot commented Feb 24, 2026

Uh oh!

rodrigouroz commented Feb 24, 2026

Uh oh!

Uh oh!

Takhoffman commented Feb 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

rodrigouroz commented Feb 24, 2026 • edited by greptile-apps bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual

Evidence

Human Verification (required)

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Greptile Summary

Confidence Score: 5/5

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

rodrigouroz Feb 24, 2026

Choose a reason for hiding this comment

Uh oh!

rodrigouroz commented Feb 24, 2026

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot commented Feb 24, 2026

Uh oh!

rodrigouroz commented Feb 24, 2026

Uh oh!

Uh oh!

Takhoffman commented Feb 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rodrigouroz commented Feb 24, 2026 •

edited by greptile-apps bot

Loading