…labs-ai#1160)

## Description

`headroom learn` ranked recommendations by a single LLM-guessed
`estimated_tokens_saved` with a flat hardcoded `confidence`, and had
**no notion of a loop**. So (1) RTK re-fetch loops were invisible - RTK
truncates a command's output, the agent re-runs larger-limit variants,
those calls *succeed* (`is_error=False`), and `analyze()` even
early-returned when a session had no failures and no events - and (2)
even when surfaced, a loop ranked no higher than a one-off mistake. This
adds loop-aware weighting plus the eval that reproduces an RTK loop,
runs it through Learn, and checks the guardrail prevents re-triggering.

Closes headroomlabs-ai#1159

## Type of Change

- [ ] Bug fix (non-breaking change that fixes an issue)
- [x] New feature (non-breaking change that adds functionality)
- [ ] Breaking change (fix or feature that would cause existing
functionality to change)
- [ ] Documentation update
- [ ] Performance improvement
- [ ] Code refactoring (no functional changes)

## Changes Made

- New `headroom/learn/loops.py`: `detect_loops()` (canonical signature
collapses RTK pagination/limit variants; classifies error vs rtk-refetch
loops; **measured** wasted tokens), `format_loops_for_digest()`,
`apply_loop_weighting()`.
- `analyzer.py`: detect loops up front (fixes the no-failure
early-return), lead the digest with them, prioritize loops in the system
prompt, re-sort after weighting.
- `models.py`: `Recommendation.is_loop_guardrail` / `loop_occurrences`.
- `benchmarks/rtk_loop_learn_eval.py` + `headroom/learn/fixtures.py`:
the two-phase RTK-loop eval and its session fixtures.
- Tests, `docs/rtk-loop-weighting.md`, CHANGELOG entry.

## Testing

- [x] Unit tests pass (`pytest`)
- [x] Linting passes (`ruff check .`)
- [ ] Type checking passes (`mypy headroom`) - not run (mypy not in my
minimal env; see Not tested)
- [x] New tests added for new functionality
- [x] Manual testing performed

### Test Output

```text
$ python -m pytest tests/test_learn/ -q
190 passed, 3 skipped, 1 warning in 5.85s
$ ruff check <changed files>
All checks passed!
```

## Real Behavior Proof

- Environment: macOS (Darwin 25.0), Python 3.10.18, fresh venv (`pip
install -e` minus the optional `hnswlib`/proxy extras, which are
unrelated to `learn`); real LLM via the analyzer's claude CLI backend
(`HEADROOM_LEARN_CLI=claude`, claude-cli 2.1.158) — no API key used.
- Exact command / steps: `HEADROOM_LEARN_CLI=claude python -c "from
benchmarks.rtk_loop_learn_eval import run_eval;
c=run_eval(use_real_llm=True); print(c.render())"`
- Observed result: the analyzer shelled out to a real model and produced
the "Commands" guardrail quoted below, naming the looping command. The
digest reports the measured 5,005-token waste and asks the model to rank
loops first, so the model emitted that figure; in this run the guardrail
ranked **headroomlabs-ai#1** and the scorecard was all-PASS (below). Caveat — real-mode
is run-dependent: the rule's wording, and whether the post-hoc
`apply_loop_weighting` fuzzy match fires, vary across runs (in one run
it did not tag the rule). The **deterministic CI eval** (stub LLM) is
the stable, reproducible artifact; this real run corroborates it.
- Not tested: the analyzer's API-key path (ANTHROPIC/OPENAI/GEMINI) —
exercised the equivalent claude CLI backend instead; `mypy`; a live
agent *obeying* the written rule end-to-end (Phase 2 is a non-recurrence
check, not a live agent — called out in the doc).

Real model output from this run, ranked headroomlabs-ai#1 at the measured 5,005-token
weight:

> **Commands** — When grepping logs (or any large file), never loop with
increasing `| head -N` limits — tool output is capped at ~4 KB
regardless of N, so repeated attempts return identical bytes. Instead:
redirect to a temp file (`grep ... > /tmp/out.txt`) then read it, or use
`grep -c` first…

```text
[PASS] loop_detected          (1 loop(s), ~5,005 tok wasted)
[PASS] guardrail_produced
[PASS] ranked_first
[PASS] names_command
[PASS] prescribes_fix
[PASS] weight_reflects_waste
[PASS] guardrail_holds
RESULT: PASS
```

(One real-mode run via the claude CLI backend. The deterministic
`pytest` eval above is the stable artifact; see the run-dependence
caveat under Observed result.)

The real run also caught an over-brittle check: an earlier
`names_command` required the literal "TimeoutError"; the real model
wrote a *more general* rule (grep + `head -N`) without it, so I fixed
the check to verify the looping **command** is named, not an incidental
literal.

## Review Readiness

- [x] I have performed a self-review
- [x] This PR is ready for human review

## Checklist

- [x] My code follows the project's style guidelines
- [x] I have performed a self-review of my code
- [x] I have commented my code, particularly in hard-to-understand areas
- [x] I have made corresponding changes to the documentation
- [x] My changes generate no new warnings
- [x] I have added tests that prove my fix is effective or that my
feature works
- [x] New and existing unit tests pass locally with my changes
- [x] I have updated the CHANGELOG.md if applicable

## Additional Notes

- No new dependencies. No network, no user/assistant content dropped —
operates on already-captured session digests.
- Kept as one logical change. mypy not run locally (minimal env); happy
to address anything CI's mypy flags.

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: JD Davis <mxjerrett@gmail.com>

🤖 I have created a release *beep* *boop*
---


<details><summary>0.28.0</summary>

##
[0.28.0](v0.27.0...v0.28.0)
(2026-06-29)


### Features

* add --disable-kompress-fallback to restore legacy PASSTHROUGH fallback
([#1185](#1185))
([f309244](f309244))
* add first-class OpenCode support (wrap, learn, mcp install)
([#559](#559))
([91cd210](91cd210))
* add HEADROOM_KEEPALIVE_EXPIRY to keep upstream connections warm
([#1124](#1124))
([85786b3](85786b3))
* **azure-foundry:** derive upstream URL from ANTHROPIC_FOUNDRY_RESOURCE
([#1138](#1138))
([e5031b0](e5031b0))
* **cache:** attribute prompt-cache misses to TTL lapse vs prefix change
([#1313](#1313))
([#1343](#1343))
([4658721](4658721))
* **code:** add Perl support to code-aware compressor
([#1125](#1125))
([f39858c](f39858c))
* headroom wrap opencode / unwrap opencode CLI
([#1105](#1105))
([b4571cc](b4571cc))
* **learn:** weight loops in Headroom Learn + RTK-loop eval
([#1160](#1160))
([14e8dc4](14e8dc4))
* **learn:** write per-project learnings to CLAUDE.local.md by default
([#1115](#1115))
([ced75e4](ced75e4))
* **proxy:** add request timeout config
([#738](#738))
([c0745d4](c0745d4))
* **proxy:** pilot hardening — inbound auth, security headers, audit
log, air-gap switch
([#1537](#1537))
([546ab55](546ab55))
* **proxy:** support glob patterns in exclude_tools
([#870](#870))
([#1259](#1259))
([a2159c0](a2159c0))
* **read-maturation:** activity-based hold-back Read maturation
(Mechanism B)
([#1068](#1068))
([723b80c](723b80c))
* **savings:** durable savings ledger + headroom savings command
([#1127](#1127))
([978ffa0](978ffa0))
* **wrap:** add --1m to preserve the 1M context window on wrap claude
([#1158](#1158))
([#1351](#1351))
([b50d9c1](b50d9c1))
* **wrap:** make tokensave the primary coding-task compressor, Serena
the backup
([#1230](#1230))
([dca9853](dca9853))


### Bug Fixes

* **agent-evals:** Phase 0 — coding-agent accuracy A/B framework
([#1037](#1037))
([84f9871](84f9871))
* **agno:** tolerate streaming tool-call SDK objects in parser
([#1312](#1312))
([#1336](#1336))
([5986c22](5986c22))
* **bedrock:** add boto3 1.41 + CRT for aws login credentials
([#1486](#1486))
([4db3bc9](4db3bc9))
* bump codebase-memory-mcp to v0.8.1
([#1284](#1284))
([530318b](530318b))
* **ccr:** make headroom_retrieve a hash-only full-content lookup
([#1532](#1532))
([c2fc4d3](c2fc4d3))
* **ccr:** propagate --no-ccr-marker flag to all compressors
([#1022](#1022))
([#1197](#1197))
([0c9b42a](0c9b42a))
* **ccr:** skip Anthropic marker emission when tool injection is
deferred
([#1273](#1273))
([2cae13d](2cae13d))
* **ci:** extend gitleaks allowlist to cover test fixtures + verified
examples
([#1539](#1539))
([d2565a6](d2565a6))
* **ci:** guarantee model present in test shards to end cache-miss
flakiness
([#1399](#1399))
([2e29c72](2e29c72))
* **ci:** normalize Windows CRLF line endings in PR governance script
([#1012](#1012))
([5194388](5194388))
* **cli:** add explicit UTF-8 encoding to file I/O in wrap commands
([#1126](#1126))
([#1164](#1164))
([a0cb798](a0cb798))
* **cli:** fall back gracefully when embedding-server sidecar is absent
([#1206](#1206))
([38f1404](38f1404))
* **cli:** harden all CLI surfaces + fix docs accuracy
([#1491](#1491))
([bd76235](bd76235))
* **cli:** wire --http2/--no-http2 (HEADROOM_HTTP2) into proxy command
([#1373](#1373))
([e06b616](e06b616))
* **cli:** wire --rpm/--tpm and HEADROOM_RPM/HEADROOM_TPM to the Click
proxy command
([#1375](#1375))
([8aab8f2](8aab8f2))
* **code:** slice tree-sitter byte offsets as UTF-8
([#1332](#1332))
([8238402](8238402))
* **code:** validate Python compressed syntax
([#1302](#1302))
([cbd361d](cbd361d))
* **code:** verify a real parse in tree-sitter availability check
([#1231](#1231))
([#1299](#1299))
([5e0bb69](5e0bb69))
* **codex:** retag threads on init so Codex Desktop history stays
visible ([#961](#961))
([#1349](#1349))
([e6bbc40](e6bbc40))
* **codex:** stop pinning Codex memory MCP to one project db
([#1269](#1269))
([ad7993b](ad7993b))
* **dashboard:** include RTK stats in the historical tab
([#1324](#1324))
([35939c3](35939c3))
* **deps:** remediate dependency CVEs and publish SBOM
([#1509](#1509))
([5771a80](5771a80))
* **docker:** persist session history across container revisions
([#1118](#1118))
([5912d65](5912d65))
* **gemini:** offload compression to the executor
([#1382](#1382))
([615848e](615848e))
* **gemini:** resolve Google model capabilities through ModelRegistry
([#1276](#1276))
([17ecad9](17ecad9))
* **install:** guard install_agent_ensure against duplicate runtime
spawns
([#1301](#1301))
([8da0b4e](8da0b4e))
* **install:** repair macOS launchd restart/start lifecycle
([#1290](#1290))
([da1a397](da1a397))
* **install:** stop duplicating ENTRYPOINT in persistent-docker runtime
command ([#833](#833))
([#1348](#1348))
([feedead](feedead))
* **io:** use UTF-8 with locale fallback and preserve line endings on
config/text I/O
([#1498](#1498))
([1baa04e](1baa04e))
* **kompress:** hard override keeps must-keep tokens regardless of model
score ([#1400](#1400))
([42612c8](42612c8))
* **langchain:** disable streaming on wrapped model during ainvoke()
([#1287](#1287))
([3590046](3590046))
* **mcp:** register managed installs with a resolvable headroom command
([#1386](#1386))
([22def93](22def93))
* **mcp:** report correct savings_percent in headroom_compress
([#1106](#1106))
([f216e43](f216e43))
* **opencode:** write local MCP config
([#1381](#1381))
([6c83790](6c83790))
* **packaging:** move hnswlib to optional [vector] extra so [all] needs
no C++ toolchain
([#1499](#1499))
([80fa086](80fa086))
* patch rtk hook script to use absolute path after register_claude_hooks
([#571](#571))
([b618d2d](b618d2d))
* **perf:** surface RTK/CLI context-tool savings in perf and the session
card ([#1433](#1433))
([9362747](9362747))
* **proxy:** add --protect-tool-results to prevent lossy compression of
exact-output Bash results
([#1374](#1374))
([51d4bcf](51d4bcf))
* **proxy:** add an Anthropic buffered read-timeout override
([#1331](#1331))
([3be2526](3be2526))
* **proxy:** add versionless Vertex AI routes for Claude Code
compatibility
([#1321](#1321))
([bb3e040](bb3e040))
* **proxy:** bind before eager preload so a hung compressor load can't
block startup
([#1500](#1500))
([d5ac07f](d5ac07f))
* **proxy:** build SSL contexts for custom CA bundles
([#1134](#1134))
([561ba17](561ba17))
* **proxy:** forward request-id headers on the streaming path
([#1100](#1100))
([#1258](#1258))
([3d59df7](3d59df7))
* **proxy:** gate CCR retrieve/compress endpoints to loopback
([#1338](#1338))
([acafb2d](acafb2d))
* **proxy:** honor force_kompress routing profile
([#996](#996))
([b4682d6](b4682d6))
* **proxy:** keep large compression results on the critical path
([#296](#296))
([#1352](#1352))
([90734b6](90734b6))
* **proxy:** offload /v1/compress to the compression executor to stop
blocking the loop
([#1501](#1501))
([27e010e](27e010e))
* **proxy:** preserve Responses memory continuations with store=false
([#1103](#1103))
([cdfeeac](cdfeeac))
* **proxy:** queue mid-turn user messages on non-Bedrock streaming path
([#1377](#1377))
([b09f027](b09f027))
* **proxy:** register interceptor in explicit transforms list when
HEADROOM_INTERCEPT_ENABLED
([#1376](#1376))
([55c700c](55c700c))
* **proxy:** report real input tokens on streaming message_start
([#1132](#1132))
([#1305](#1305))
([70cc96a](70cc96a))
* **proxy:** retry upstream 429 with Retry-After on both forwarders
([#1329](#1329))
([90bee89](90bee89))
* **proxy:** retry upstream 529 overloaded like 429 on both forwarders
([#1495](#1495))
([547b15d](547b15d))
* **proxy:** stop re-compressing headroom_retrieve output and emitting
unredeemable markers
([#1323](#1323))
([43494ff](43494ff))
* **proxy:** strip Codex lite header from OpenAI WebSockets
([#1543](#1543))
([5d3803a](5d3803a))
* **read-lifecycle:** persist STALE Read originals in the CCR store
([#1488](#1488))
([9157173](9157173))
* recover persistent proxy feature checks and reject non-Copilot
exchange URL
([#1465](#1465))
([16c638b](16c638b))
* remove agents.md
([#1540](#1540))
([a7d3360](a7d3360))
* respect COPILOT_PROVIDER_TYPE env var when provider_type is auto
([#549](#549))
([24cf256](24cf256))
* restore token-mode compression on frozen prefixes
([#1489](#1489))
([8e0dadf](8e0dadf))
* **router:** degrade to pure-Python detection on native panic
([#1123](#1123))
([#1260](#1260))
([a00fb67](a00fb67))
* **rtk:** stop hook registration timing out on a forked daemon
([#1314](#1314))
([9758817](9758817))
* **smart-crusher:** honor enable_ccr_marker on the opaque-blob path
([#1130](#1130))
([27d6f8e](27d6f8e))
* **subscription:** only reset 5h contribution on real rollover, not API
jitter
([#1255](#1255))
([8d6c175](8d6c175))
* **subscription:** run transcript token scan off the event loop
([#1263](#1263))
([f03021f](f03021f))
* surface output reduction without a restart, and explain $0.00 savings
on Python 3.14
([#1296](#1296))
([c30ec4c](c30ec4c))
* **tests:** reset whole headroom logger subtree so caplog stays
deterministic
([#1117](#1117))
([fda4670](fda4670))
* **tls:** add HEADROOM_TLS_STRICT=0 toggle for corporate SSL inspection
([#1308](#1308))
([#1341](#1341))
([52068dd](52068dd))
* **tokenizers:** price CJK/Kana/Hangul at ~1 token per char in
EstimatingTokenCounter
([#1093](#1093))
([a35fe86](a35fe86))
* **transforms:** gate tool string output from lossy compression
([#1307](#1307))
([#1387](#1387))
([c6c921a](c6c921a))
* **websocket:** harden responses websocket origin handling
([#1481](#1481))
([c632023](c632023))
* **windows:** pin UTF-8 encoding on text-mode subprocess calls
([#1311](#1311))
([d633e81](d633e81))
* **wrap:** add Copilot unwrap command
([#1251](#1251))
([b4fde0c](b4fde0c))
* **wrap:** isolate proxy stdio from proxy.log on Windows
([#1191](#1191))
([959ab0d](959ab0d))
* **wrap:** keep agent savings opt-in
([#1294](#1294))
([b829ceb](b829ceb))
* **wrap:** show the dashboard URL when the proxy is already running
([#1313](#1313))
([b0146c4](b0146c4))


### Performance Improvements

* **compression:** take large cold-start contexts off the synchronous
kompress path
([#1171](#1171))
([#1298](#1298))
([6c68ff4](6c68ff4))
</details>

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>