v1.9.0

What's Changed

• Fix audit and OneNote help after b23 review by @wu-changxing in #1935
• co ai opens on a card: invite code for the owner, skills folded, directory opt-in (#1939) by @wu-changxing in #1940
• The README introduces co rem, your agent's memory, as it is today: co wiki, Experimental, renamed in 1.9.0 by @wu-changxing in #1945
• Project pages from your own Codex and Claude Code messages, recent projects first (#1943 stage 2) by @wu-changxing in #1947
• co wiki init is one command with value in the first minutes: your page printed, then written by itself, then your recent projects (#1943) by @wu-changxing in #1946
• 1.9.0a1: co rem investigation searches evidence files, Skills carry rules only (≤15k per turn), plus fixes found measuring it by @wu-changxing in #1942
• Sessions typed in a workspace are filed under the repository they worked in, and recent folders get a page (#1943, stage 2b) by @wu-changxing in #1948
• People are investigated recent correspondents first, from new mail only, and the daily round splits into an unfinished portion and runs that follow what is new (#1943 stage 3) by @wu-changxing in #1951
• co rem 1.9.0: addresses held for review, quiet nights cost nothing, the model's tier is measured (#1844, #1846, #1847) by @wu-changxing in #1949
• 1.9.0a2: the wiki is renamed co rem, and co wiki notebooks, schedules and scripts carry over (#1932) by @wu-changxing in #1953
• A co rem help page moves nothing; 1.9.0a2 notes name the first-run work by @wu-changxing in #1961
• A project page lists a worktree as its main checkout, first, so investigation reads the project and not a stale agent's copy (#1955) by @wu-changxing in #1965
• co rem sync names the daily cap where you look, and --all stays under it (#1957) by @wu-changxing in #1969
• co rem: six fixes from the 1.9.0a1 real-notebook run (cited handles, CLI reference, task copies, schedule, sizes, status line) by @wu-changxing in #1970
• co rem: one prompt per subject, the gathered material as the only source, only what is new; maintenance keeps pages near 15k (#1956) by @wu-changxing in #1971
• An org page asks the mailbox for its domain and reads the init archive (#1963) by @wu-changxing in #1967
• Prepare 1.9.0a3: 1.9.0a2, published by @wu-changxing in #1973
• co rem: one catalog page per skill, one project page per repository, older split pages merged as aliases (#1974, 4 and 5) by @wu-changxing in #1976
• co rem: no stamp without material, one people queue with the owner first, one bad citation drops one line, and the daily round says which mailboxes it reads (#1974 B) by @wu-changxing in #1977
• co rem reads what the owner typed in Codex Desktop (#1978) by @wu-changxing in #1980
• co rem page skills: a person page opens on its lead, no coverage filler, project pages from messages say once what they cannot know (#1974 6, 7, 8) by @wu-changxing in #1975
• co rem: org investigations work again: Outlook is asked for the domain's first label, and one mailbox failing no longer ends the page (#1981) by @wu-changxing in #1991
• co rem: extracted project notes reach the page they name, and a written page starts the next window (#1985, #1983) by @wu-changxing in #1993
• co rem: no model turn for an empty since-window, a stale checkout named as stale, and services off the people map (#1984, #1982, #1987) by @wu-changxing in #1994
• Prepare 1.9.0a4: what the real notebook found by @wu-changxing in #1995
• 1.9.0a5: one look for every co command and co audit checks it; co rem status is a dashboard; the first run writes your page, 3 people and projects (#1997, #1996) by @wu-changxing in #2004
• urllib3 floor and lock to 2.8.0: three new advisories against 2.7.0 fail every PR's lockfile audit by @wu-changxing in #2007
• co rem: maintaining extraction notes carries no source Skill, and no turn is pointed at the rationale docs (#2000, #2001) by @wu-changxing in #2006
• co rem: the reader and status count from one census, status tells the truth, and an upgraded notebook is tidied (#2008 5-7, #1999) by @wu-changxing in #2010
• The look across co is one look, and co audit sees what a terminal shows: no [env] noise, co status on the palette, no auto-highlighting, doctor the same in a terminal and a pipe (#2008) by @wu-changxing in #2009
• co rem first run: capped, one estimate from real runs; your own page has its own spec and name; a written page may not say 'not investigated yet' (#2008 A) by @wu-changxing in #2011
• 1.9.0a5 notes: what the real-data run fixed before the tag by @wu-changxing in #2012
• co rem: a Gmail date without a timezone no longer stops investigate me, and maintenance can add to a page again (#2013, #2014) by @wu-changxing in #2020
• co rem: mail a page already cites is not new material, and run records say how long each stage took (#2015, #2016) by @wu-changxing in #2022
• co rem: tidy runs again after the map, services and mail providers make no pages, the owner's page is rewritten, and a page may not grow past 20k (#2017, #2018, #2019) by @wu-changxing in #2021
• Prepare 1.9.0a6: what 1.9.0a5's acceptance run found by @wu-changxing in #2023
• A refused project page waits for newer messages, is told its size first, and runs report themselves (#2026, #2029, #2030, #2033, #2034) by @wu-changxing in #2037
• co rem's daily cap is 50, the owner's choice (#2032) by @wu-changxing in #2036
• co rem: coding agents are the owner's tools, folded addresses are cited, a company writing as itself is a service, and a stalled mail archive is reported and resumed (#2027, #2028, #2031, #2035) by @wu-changxing in #2038
• Prepare 1.9.0a7: what 1.9.0a6's acceptance run found by @wu-changxing in #2039
• Every page is told the size limit before its turn, a refused investigation waits for new mail, and the queue ranks what sync wrote (#2041, #2046, #2045) by @wu-changxing in #2047
• co rem: investigations read a partial mail archive, every run has a usage stage, sync ends on a summary, and the investigation budget moves (#2042, #2043, #2044, #1990) by @wu-changxing in #2052
• Prepare 1.9.0a8: what 1.9.0a7's acceptance run found by @wu-changxing in #2053
• A citation jumps to its source, a link to a missing page says so, and tidy unlinks what it archived (#2054) by @wu-changxing in #2055
• Lines about the run, not the subject, are removed from pages: 'web: not searched; Wiki runs are offline' (#2058) by @wu-changxing in #2061
• co rem's terminal output gets a designed layout: today first, one value column, glyphs with meaning, and fixes under what they fix by @wu-changxing in #2070
• A person page is for someone the owner corresponds with; tidy archives the 263 one-way templates (#2057) by @wu-changxing in #2062
• co rem keeps an index of its map and raw material in SQLite beside the pages, and People opens as a table (#2067) by @wu-changxing in #2073
• co rem reader: opens on the night, People as a sheet, pages open on what is owed and their cited facts (#2064, #2065, #2068) by @wu-changxing in #2077
• co rem reader: the People sheet fits 1440 with What's open beside the name, and pages about "the user" read as "you" by @wu-changxing in #2085
• History is eight dated milestones on a timeline, and the map's mail count moves to the lead (#2059) by @wu-changxing in #2069
• The first mention of a person the notebook has a page for links to it (#2060) by @wu-changxing in #2074
• Project pages know the user owns their own projects and read the README; every page in English (#2060) by @wu-changxing in #2076
• Evidence is a month of one mailbox per file: Ody's page 2.77M → 1.48M input tokens (#2080) by @wu-changxing in #2083
• Prepare 1.9.0a9: the notebook, designed by @wu-changxing in #2089
• Every page opens on cited Facts read from the material before the prose, and an Insight section that says something (#2068) by @wu-changxing in #2081
• The first run writes the busiest projects, never a private journal unasked; a one-off Codex task folder is not a project (#2079) by @wu-changxing in #2090
• The owner's possible own addresses are asked about without person pages of their own (#2078) by @wu-changxing in #2091
• Commands can read a value saved with co env set --secret by @wu-changxing in #2082
• co linear: Linear issues from the terminal, on Linear's GraphQL API (#2049) by @wu-changxing in #2063
• co slack: read channels, history, threads and search on Slack's Web API; co auth slack from one manifest (#2051) by @wu-changxing in #2075
• co canny: our own Canny commands on Canny's REST API (#2050) by @wu-changxing in #2072
• README: native connection wall (real text, one SVG per logo) + Slack by @wu-changxing in #2056
• Prepare 1.9.0a10 REM preview by @wu-changxing in #2093
• Make REM's morning context the reader's first screen by @wu-changxing in #2095
• Prepare 1.9.0a11 REM morning preview by @wu-changxing in #2100
• Make REM reader snapshots safe on Windows by @wu-changxing in #2101
• Prepare 1.9.0a12 REM Windows preview by @wu-changxing in #2102
• docs: audit REM record pages and interaction by @wu-changxing in #2108
• feat(rem): connected memory reader preview 1.9.0a13 by @wu-changxing in #2110
• feat(rem): investigate recent memory on first run (1.9.0a14) by @wu-changxing in #2119
• fix(rem): keep first-run projects within the map (1.9.0a15) by @wu-changxing in #2120
• Fix PR review action JSON envelope validation by @wu-changxing in #2125
• fix(rem): bound Gmail scans on full windows by @wu-changxing in #2126
• Ground REM first-run project insight in local evidence (1.9.0a17) by @wu-changxing in #2123
• docs: README as a short, trustworthy front page; add SECURITY.md and CODE_OF_CONDUCT.md by @wu-changxing in #2099
• README: lead with the idea — one CLI for all the context, why a CLI and not MCP; co mcp coming by @wu-changxing in #2129
• REM owner init surfaces cited decisions and next steps by @wu-changxing in #2132
• Complete co rem branding and release 1.9.0a19 by @wu-changxing in #2148
• Finish co rem reader labels and prepare 1.9.0a20 by @wu-changxing in #2160
• fix(ci): Warn when blog story judge has no credits by @wu-changxing in #2170
• docs: showcase co rem on the GitHub homepage by @wu-changxing in #2192
• docs: restore the visible homepage logo wall by @wu-changxing in #2193
• Improve REM first-run coverage, citations and full-memory reader by @wu-changxing in #2141
• docs: present co rem as decentralized context flow by @wu-changxing in #2194
• REM a22: source-backed first-run project findings by @wu-changxing in #2196
• [rem] Clarify mapped contact dates in the reader by @wu-changxing in #2204
• [rem] Add co rem merge command to fold duplicate pages and record aliases by @wu-changxing in #2208
• Cancel superseded PR test runs by @wu-changxing in #2175
• [rem] Show first map results and fix large-reader trust issues by @wu-changxing in #2213
• [ci] Run the full CLI help audit once by @wu-changxing in #2215
• [rem] Open written notebooks faster and pin preview docs by @wu-changxing in #2217
• [rem] Clarify status when memories exist but updates are off by @wu-changxing in #2219
• Release 1.9.0a26: faster written notebooks by @wu-changxing in #2220
• [rem] Filter institutional service desks and enforce project entry gate (#2201, #2202) by @wu-changxing in #2221
• [docs] Match published a26 release copy by @wu-changxing in #2222
• [rem] Explain service exclusions in People counts by @wu-changxing in #2224
• [release] 1.9.0a28 preview (context over control in investigation) by @wu-changxing in #2226
• [rem] Keep complete cited mail available in the reader by @wu-changxing in #2225
• [rem] Put mapped-page action within phone reach (1.9.0a29) by @wu-changxing in #2228
• [release] 1.9.0a30 preview (full-cohort concurrent onboarding & investigation) by @wu-changxing in #2244
• fix(rem): respect budget gate and handle missing resets_at in quota points by @wu-changxing in #2247
• [rem] Audit Project claims before promotion (1.9.0a31) by @wu-changxing in #2238
• [rem] Keep Skill claims tied to inspectable outputs (1.9.0a32) by @wu-changxing in #2252
• [rem] Keep first contact evidence based in a33 preview by @wu-changxing in #2262
• [rem] Reuse scoped session inputs across CLI runs (1.9.0a34) by @wu-changxing in #2267
• [rem] Keep phone source actions in the first fold (1.9.0a35) by @wu-changxing in #2269
• fix(rem): preserve Outlook HTML source boundaries by @wu-changxing in #2271
• Keep REM Project purpose and evidence reachable in the first screen by @wu-changxing in #2272
• Find passages in long REM source dialogs by @wu-changxing in #2274
• [rem] Put Skill finding sources beside the lead (1.9.0a39) by @wu-changxing in #2275
• [rem] Clarify Skill usage and phone actions (1.9.0a40) by @wu-changxing in #2277
• Map historical REM contacts and start nightly upkeep from init by @wu-changxing in #2286
• Fix REM Gmail historical metadata transport by @wu-changxing in #2288
• co rem on claude-code: default to Sonnet 5.5, and fresh headless turns no longer lock their folder by @wu-changxing in #2294
• Prepare REM history preview 1.9.0a41 by @wu-changxing in #2289
• Bring REM historical contacts into the first People view by @wu-changxing in #2291
• Keep REM people found before a historical mail failure by @wu-changxing in #2299
• Prepare REM preview 1.9.0a42 with retained discovery and reachable contacts by @wu-changxing in #2300
• Show failed REM investigations on Home by @wu-changxing in #2303
• docs: Explain nightly upkeep in the REM init overview by @wu-changxing in #2304
• Prepare REM preview 1.9.0a43 by @wu-changxing in #2305
• Release co rem 1.9.0 by @wu-changxing in #2311

Full Changelog: v1.8.9...v1.9.0