rossoctl/context-guru

View on GitHub

Context engineering manages an agentic system's context: persisting and efficiently retrieving conversations and trajectories for the purpose of optimizing LLM context, this may include compacting and summarizing context to reduce costs and latency while preserving accuracy.

Stars: 54
Forks: 22
Open beginner issues: 0
Indexed issues: 62
Avg merge: 20h 42m
Merged PRs (30d): 89

Dominant language: Go
License: Apache-2.0
Last GitHub push: Sep 18, 2026
Latest indexed: Sep 20, 2026
Contributing guide: Contributing guide
Code of conduct: Code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

62 open issues indexed

Tool-output dedup marker is indistinguishable from genuine unchanged output right after an edit Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#281 · Opened Sep 19, 2026
cache_state: pre_expiry has no working configuration — cache_aware_summarizer ignores the key, and cannot run on Anthropic traffic at all Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

rossoctl/context-guru#275 · Opened Sep 17, 2026
docs: ~~strikethrough~~ renders as literal tildes on two built pages (pymdownx.tilde absent), and enabling it would corrupt four ~figures Open

Difficulty 2/5 1-3 hours Newbie friendliness 88/100

rossoctl/context-guru#272 · Opened Sep 16, 2026
LOCA rig: readout resolved passes by a glob that cannot match, and the obvious fix silently aliases both arms Open

benchmark_specific bug priority: low

Difficulty 3/5 1-2 days Newbie friendliness 25/100

rossoctl/context-guru#269 · Opened Sep 16, 2026
LOCA rig: context clearing is ENABLED but has never fired, in any pass, even 30k over its trigger Open

benchmark_specific bug priority: low

Difficulty 4/5 3-5 days Newbie friendliness 45/100

rossoctl/context-guru#268 · Opened Sep 16, 2026
LOCA rig: 25% of sonnet responses truncate at max_tokens 4096, corrupting the next tool call Open

benchmark_specific bug priority: low

Difficulty 4/5 3-5 days Newbie friendliness 35/100

rossoctl/context-guru#267 · Opened Sep 16, 2026
Measure extract_llm_sweep on SWE-bench: does a removal pay when S and T are not thin? Open

enhancement

Difficulty 4/5 3-5 days Newbie friendliness 38/100

rossoctl/context-guru#266 · Opened Sep 16, 2026
cachesplit's -34.1% result is described as general, but it only applies to Claude Code CLI traffic Open

documentation priority: low

Difficulty 2/5 1-3 hours Newbie friendliness 86/100

rossoctl/context-guru#263 · Opened Sep 16, 2026
test(offload): TestExtractLLMDoesNotBookASingleFlightFollowersSaving passes and fails in the same CI job on one commit Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

rossoctl/context-guru#253 · Opened Sep 15, 2026
plugin: the proxy log sits outside CONTEXT_GURU_STATE, so a failure report can show another run's lines (and its base URL) Open

Difficulty 3/5 1-2 days Newbie friendliness 74/100

rossoctl/context-guru#251 · Opened Sep 15, 2026
Split Trigger so extract/extract_llm cannot accept cache_state keys they ignore Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

rossoctl/context-guru#247 · 1 comment · Opened Sep 14, 2026
Seven components/offload tests fail on main at -count=2: extract_econ global state leaks between runs Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

rossoctl/context-guru#246 · Opened Sep 14, 2026
summarize's cache gate is wrong on OpenAI-shaped backends in both cache modes, in opposite directions Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#245 · 1 comment · Opened Sep 14, 2026
The store's pin budget saturates, silently, and no metric says so Open

Difficulty 3/5 1-2 days Newbie friendliness 55/100

rossoctl/context-guru#244 · Opened Sep 14, 2026
adjudication prompt: "keep everything is a valid and often correct answer" may be licensing a 2% drop rate — needs offline measurement, not an edit Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

rossoctl/context-guru#242 · 1 comment · Opened Sep 14, 2026
Offer provider-native compaction as an alternative to summarize, for deployments with no summarizer model Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#241 · Opened Sep 14, 2026
saved_usd prices our own token count at the provider's rate, under-reporting savings by ~12% (measured) Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

rossoctl/context-guru#240 · 1 comment · Opened Sep 13, 2026
summarize: learn the client's own compaction point instead of assuming a fixed fraction of the window Open

Difficulty 5/5 Over a week Newbie friendliness 38/100

rossoctl/context-guru#239 · 1 comment · Opened Sep 13, 2026
summarize: extend an existing summary instead of discarding it when the covered span moves Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#238 · Opened Sep 11, 2026
context-guru-plugin: two install.sh tests fail on any machine with context-guru-proxy already on PATH Open

Difficulty 2/5 1-3 hours Newbie friendliness 84/100

rossoctl/context-guru#235 · Opened Sep 10, 2026
econ trigger: the turn horizon ignores its own removal, and swings two orders of magnitude on transcript shape Open

bug

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#232 · Opened Sep 9, 2026
econ trigger: the approval estimate is a curve in batch size, modelled as a scalar — and it sticks at its floor Open

bug

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#230 · 1 comment · Opened Sep 9, 2026
extract_llm_sweep: should the econ trigger have a context-pressure floor before it pays to ask? Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#229 · Opened Sep 8, 2026
plugin installs inherit server-shaped dashboard persistence defaults (7-day, 512MB on-disk) that a single-user test install does not need Open

design

Difficulty 4/5 3-5 days Newbie friendliness 45/100

rossoctl/context-guru#228 · Opened Sep 8, 2026
extract_llm: a saturated reserve stops the ratio tracker learning, so exploration never self-terminates and re-pays its trial calls on every session Open

Difficulty 5/5 Over a week Newbie friendliness 28/100

rossoctl/context-guru#214 · Opened Sep 6, 2026
proxy: TestFeedbackMailCannotInjectHeaders flakes because the mail goroutine reads CG_MAIL_DEV_SINK after its test restored it Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

rossoctl/context-guru#213 · Opened Sep 6, 2026
offload: summarize's kept-verbatim trim can force a re-summary, and that cache-write is not counted by expand_prefix_flips Open

Difficulty 4/5 3-5 days Newbie friendliness 42/100

rossoctl/context-guru#209 · 3 comments · Opened Sep 4, 2026
proxy: usage_unreadable is structurally unreachable for a streamed response, which is where a spliced window is most likely to hide the block Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

rossoctl/context-guru#207 · Opened Sep 4, 2026
test(proxy): TestArriveCancelsAndReports waits on a counter published before the state it asserts Open

Difficulty 2/5 1-3 hours Newbie friendliness 78/100

rossoctl/context-guru#206 · 2 comments · Opened Sep 3, 2026
expand: one call retrieves one id and nothing tells the model it may batch; consider an ids list and programmatic (selective) retrieval Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#202 · Opened Sep 3, 2026
offload: the replay depth-gate bypass is justified per-message but keyed per-content, so a replay can splice at depth into uncompacted content Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

rossoctl/context-guru#199 · 1 comment · Opened Sep 3, 2026
test: the replay-branch Irreversible table has no completeness check, unlike gateExempt for components Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

rossoctl/context-guru#198 · Opened Sep 3, 2026
offload: extract_llm and extract_llm_sweep share the cg:res: namespace with no component tag, so a depth-gate bypass can rest on another component's decision Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

rossoctl/context-guru#197 · 3 comments · Opened Sep 3, 2026
test: proxy package cannot run at -count>1 — two tests assert on process-global metrics counters Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

rossoctl/context-guru#192 · 3 comments · Opened Sep 3, 2026
observability: the cg_stash_* family has no Grafana panel and no alert, including the one counter that reports a broken reversibility promise Open

Difficulty 3/5 1-2 days Newbie friendliness 76/100

rossoctl/context-guru#189 · Opened Sep 3, 2026
decision(extract_llm_sweep): should a locally-refused ErrNoPrefix ask count as a call? #178 says no, which steps down a published counter Open

rossoctl/context-guru#182 · 1 assignee · Opened Sep 2, 2026
refactor(cheapmodel): Usage() is process-global and its comment still asserts the premise that produced #176 Open

Difficulty 5/5 Over a week Newbie friendliness 45/100

rossoctl/context-guru#181 · Opened Sep 2, 2026
feat(promexport): cg_extract_* has no component label, so an alert cannot say which component went underwater — and now inherits cost_source without expressing it Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

rossoctl/context-guru#180 · Opened Sep 2, 2026
fix(dash): /api/components counts free frozen replays as acts, so act_rate reads as work the component paid for Open

Difficulty 3/5 1-2 days Newbie friendliness 74/100

rossoctl/context-guru#179 · Opened Sep 2, 2026
fix(offload): kept-verbatim marks are global, so one session's expand exempts that content for every session Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

rossoctl/context-guru#175 · Opened Sep 1, 2026
cg SSE peek: an adjudicate leak is counted as an expand leak in sse_expand_after_stream Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

rossoctl/context-guru#174 · Opened Sep 1, 2026
dash: checkpoint() discards wal_checkpoint's busy result, so #115's WAL protection silently no-ops under load Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

rossoctl/context-guru#166 · Opened Sep 1, 2026
sweep_quote_fabricated swings 3x between runs and inverts between arms, so it cannot gate a PR Open

Difficulty 5/5 Over a week Newbie friendliness 45/100

rossoctl/context-guru#165 · Opened Sep 1, 2026
extract_llm_sweep: empty and wrong-tool prefix-ask replies are both filed as sweep_unparseable Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

rossoctl/context-guru#164 · Opened Sep 1, 2026
test: TestCtlGetCampaignAggregatesPredictedAndRealPerTenant is load-sensitive — real_saved_usd reads 0 on a starved CI runner Open

Difficulty 3/5 1-2 days Newbie friendliness 55/100

rossoctl/context-guru#163 · Opened Sep 1, 2026
test: TestConcurrentCallsDoNotRaceOnTheGateHistogram fails intermittently under full-suite load (the race it needs is never exercised) Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

rossoctl/context-guru#162 · Opened Sep 1, 2026
dash: the first savings bar renders 158% in a 100%-wide track — a ratio presented as a share Open

Difficulty 2/5 1-3 hours Newbie friendliness 78/100

rossoctl/context-guru#158 · Opened Sep 1, 2026
fix(metrics): 14 Snapshot counters reach /stats but never Prometheus, including every summarize/agentdiet failure signal Open

Difficulty 4/5 3-5 days Newbie friendliness 68/100

rossoctl/context-guru#156 · Opened Sep 1, 2026
docs: searchfold and linecap are the only components with no page Open

Difficulty 2/5 1-3 hours Newbie friendliness 88/100

rossoctl/context-guru#149 · Opened Sep 1, 2026
docs: internal contextguru.vpc.cloud9.ibm.com hostname linked from 2 public pages Open

Difficulty 2/5 1-3 hours Newbie friendliness 78/100

rossoctl/context-guru#148 · Opened Sep 1, 2026

rossoctl/context-guru

Get new issues in your inbox