@damian87/omp 0.10.0 → 0.13.0

package/.github/copilot-instructions.md

@@ -13,6 +13,22 @@ Default behaviours installed with this repo. Override per project as needed.
 - Read the diff before committing.
 - If unsure about scope, ask.
 
+## Cost/token discipline
+Cost data is local, best-effort, and estimated. `omp cost [--today] [--session <id>]`
+summarizes prompt/tool token estimates from the hook ledger; it is not provider billing.
+
+The cost hooks apply when this plugin's `hooks/hooks.json` is active in a Copilot CLI
+session. They give session-wide visibility for skills invoked inside that session, not
+standalone coverage for copied skills, raw shell scripts, or external CLIs.
+
+Before rerunning noisy commands or failed edits, inspect the latest output and narrow the
+next attempt. Prefer bounded summaries for large logs. Oversized postToolUse output is
+minimized before it re-enters model context, with raw output preserved on disk and savings
+recorded in the cost ledger. Diagnostics (errors, stack traces, assertions) are preserved
+inline; other trimmed detail must be recovered by re-reading the raw file path noted in the
+hook output (an extra tool call), so full fidelity depends on the model following that pointer.
+Budget gates and retry-cost guidance are not current live behavior.
+
 ## Skills
 Slash commands under `.github/skills/<name>/SKILL.md` are auto-discovered by Copilot. See `omp list` for the catalog active in this project.
 

package/.github/skills/jira-ticket/SKILL.md

@@ -18,21 +18,21 @@ Use `/jira-ticket` when work tracking is requested.
 ### Create
 - Build from an approved plan or implementation slice (a plan file or markdown)
 - Include: Summary, Description, Acceptance Criteria
-- Render the payload with `omp jira render <plan-file>` — this **never** writes to Jira. To create/comment/update, run `omp jira apply <plan-file-or-ticket-key>`, which defaults to **dry-run** and only writes when Jira is configured with `JIRA_MODE=live` and the user has explicitly confirmed.
+- Render the payload with `omp jira render <plan-file>` — this **never** writes to Jira. To create/comment/update, run `omp jira apply <plan-file-or-ticket-key>`, which defaults to **dry-run** and only writes when Jira is configured with `JIRA_MODE=live`, the user has explicitly confirmed, and the CLI has explicit body/field support for the operation you need.
 
 ### Comment
 - Add implementation evidence, verification results, or status updates
 - Format for readability (use Jira wiki markup, not Markdown)
-- Apply with `omp jira apply <ticket-key> --comment` (dry-run by default; preview before confirming)
+- Draft the comment payload and preview via dry-run only; do not run live comment writes from this skill until the CLI accepts an explicit comment body/body file.
 
 ### Safe update
 - Only update known simple fields (summary, description, labels)
-- Apply with `omp jira apply <ticket-key> --update` (dry-run by default)
+- Draft update fields and preview via dry-run only; do not run live update writes from this skill until the CLI accepts explicit update fields.
 - Do not guess transitions, issue links, project keys, or secrets
 
 ## Rules
 
-- Always preview with `omp jira render` or `omp jira apply … --dry-run` (the default) before any live write
+- Always preview with `omp jira render` or `omp jira apply … --dry-run` (the default); live writes require explicit user confirmation plus CLI support for the exact payload being written
 - If Jira config is missing, the commands stay in dry-run and print the payload — never fail silently
 - Do not guess project keys, transitions, or credentials
 - Keep acceptance criteria testable and specific

package/.github/skills/omp-autopilot/SKILL.md

@@ -73,3 +73,7 @@ Before claiming done:
 - [ ] Lint clean (if applicable)
 - [ ] `/verify` or `/ultraqa` produced PASS evidence
 - [ ] No uncommitted work left behind
+
+## Cost/token note
+
+This skill can drive multiple tool calls or long-running output. Use `omp cost [--today] [--session <id>]` for local hook-ledger estimates only; it is not provider billing. Keep injected summaries concise and prefer bounded output when rerunning noisy commands.

package/.github/skills/research-codebase/SKILL.md

@@ -56,3 +56,7 @@ Show concise summary to user with key file references. Ask if they have follow-u
 ### 7. Follow-ups
 
 If the user has follow-ups, read `reference/follow-up.md` for the append protocol.
+
+## Cost/token note
+
+This skill can drive multiple tool calls or long-running output. Use `omp cost [--today] [--session <id>]` for local hook-ledger estimates only; it is not provider billing. Keep injected summaries concise and prefer bounded output when rerunning noisy commands.

package/.github/skills/schedule/SKILL.md

@@ -69,3 +69,7 @@ omp schedule add --id nightly-tests --cron "0 2 * * *" \
   --prompt "Run the test suite; if anything fails, open an issue with the log." \
   --allow-all-tools --max-runs 7
 ```
+
+## Cost/token note
+
+This skill can drive multiple tool calls or long-running output. Use `omp cost [--today] [--session <id>]` for local hook-ledger estimates only; it is not provider billing. Keep injected summaries concise and prefer bounded output when rerunning noisy commands.

package/.github/skills/team/SKILL.md

@@ -127,3 +127,7 @@ Use `/ralplan` before `/team` to produce the plan. Use `/verify` after completio
 - Each pane is an independent session — no shared state
 - Workers can message each other via `omp team api send-message` (runtime mode only)
 - If tasks depend on each other, use `/ralph` instead
+
+## Cost/token note
+
+This skill can drive multiple tool calls or long-running output. Use `omp cost [--today] [--session <id>]` for local hook-ledger estimates only; it is not provider billing. Keep injected summaries concise and prefer bounded output when rerunning noisy commands.

package/.github/skills/ultrawork/SKILL.md

@@ -71,3 +71,7 @@ Summarise: completed, failed, blocked. Then clear the tracked state with `omp ul
 - `Completed` — items done with evidence
 - `Failed/blockers` — items that couldn't be completed and why
 - `Verification` — test/lint/build results per wave
+
+## Cost/token note
+
+This skill can drive multiple tool calls or long-running output. Use `omp cost [--today] [--session <id>]` for local hook-ledger estimates only; it is not provider billing. Keep injected summaries concise and prefer bounded output when rerunning noisy commands.

package/.github/skills/weighted-consensus/SKILL.md

@@ -127,3 +127,7 @@ default roster is used.
 
 Pair with `/code-review` (pass a review rubric + the diff as context) for a
 multi-model review verdict, or with `/ralplan` to weigh competing plans.
+
+## Cost/token note
+
+This skill can drive multiple tool calls or long-running output. Use `omp cost [--today] [--session <id>]` for local hook-ledger estimates only; it is not provider billing. Keep injected summaries concise and prefer bounded output when rerunning noisy commands.

package/README.md

@@ -55,7 +55,7 @@ That's it.
 - **Persistent execution** — Ralph, UltraQA, and Ultrawork keep going until the goal is verified
 - **File-state coordination** — workers swap typed messages over an outbox/inbox cursor with atomic `O_EXCL` task locks; no broker or daemon to babysit
 - **Chat bridge** — `omp gateway` runs long-lived chat connectors (Slack today, more next) so you can DM Copilot from anywhere
-- **Lifecycle hooks** — `SessionStart`, `UserPromptSubmit`, `PreToolUse`, `PostToolUse`, `SessionEnd`, `Error`
+- **Lifecycle hooks** — `sessionStart`, `userPromptSubmitted`, `preToolUse`, `postToolUse`, `postToolUseFailure`, `sessionEnd`, `errorOccurred`
 - **Doctor included** — `omp doctor` verifies plugin manifest, skills discovery, hooks, and the underlying `copilot` CLI in one shot
 
 ---
@@ -84,6 +84,7 @@ That's it.
 
 - **Context & history as CLI subcommands** — `omp state` (key-value with TTL), `omp project-memory` (notes + directives), `omp trace` (per-session timeline + summary), `omp goal` / `omp memory sync` (managed repo context), `omp daily-log`
 - **Lightweight Copilot context** — managed instructions keep only the repo goal plus on-demand memory commands; set `OMP_DISABLE_INSTRUCTIONS_MEMORY=1` to skip writing the managed block entirely
+- **Estimated cost ledger** — `omp cost [--today] [--session <id>]` summarizes local prompt/tool token estimates recorded by hooks. These are best-effort estimates, not provider billing.
 - **File-state worker coordination** — outbox JSONL + byte cursor, atomic `O_EXCL` task locks, optimistic CAS on claim
 - **Idle nudge** — content-based pane idle detection that pokes stuck workers
 - **Mode-state loops** — single source of truth per loop (Ralph/Ultrawork/UltraQA state files)
@@ -195,6 +196,7 @@ omp ralph start "<task>" [--max-iterations N]
 omp ultrawork start "<objective>" [--task-count N]
 omp ultraqa start "<goal>" [--max-cycles N]
 omp council "<question>" [--models a,b,c] [--context @file] [--json]   # multi-model council
+omp cost [--today] [--session <id>]            # summarize estimated hook-ledger tokens
 omp comms status | send | recv | ask        # drive a running copilot tmux session
 omp gateway serve [--only slack]            # run chat connectors (today: slack)
 omp gateway status [--json]                 # per-connector readiness (no sockets)
@@ -272,6 +274,7 @@ omp grows in vertical slices. Items aren't pinned to specific semver versions
 - **Slack outbound — `omp gateway notify`** — stateless REST `chat.postMessage` from any process (cron `--notify-target`, in-session `/slack <message>`, ad-hoc `omp gateway notify --text "..."`). Default destination from `SLACK_HOME_CHANNEL`; explicit `--target slack:C…/G…/D…/U…` overrides; `U…` auto-resolves to a DM via `conversations.open`.
 - **Weighted-consensus council** — multi-model council with role weights + minority report. Via `omp council` or `/weighted-consensus`.
 - **Suggest** — `omp suggest "<task>"` recommends a slash-skill workflow without launching one.
+- **Estimated cost ledger** — hook-driven prompt/tool token estimates are visible through `omp cost`; oversized `postToolUse` output is minimized before it re-enters model context, with raw output preserved on disk (diagnostics kept inline; other trimmed detail is recoverable by re-reading the raw path the hook reports — an extra tool call, so best with capable models). Budget gates and retry-cost guidance remain next-step optimization work, not shipped behavior yet.
 
 ### Up next
 

package/dist/src/cli.js

@@ -26,7 +26,7 @@ function printResult(result, json) {
     }
 }
 function help() {
-    return `oh-my-copilot\n\nRun \`omp\` with no arguments to launch copilot (permissions bypass OFF).\nUse \`omp help\` to show this list.\n\nCommands:\n  (no args)                                     launch copilot (bypass OFF by default)\n  version [--json]\n  list [--json]\n  setup [--dry-run] [--scope project|user] [--plugin-root <dir>] [--json]\n  doctor [--json] [--copilot-bin <path>] [--skip-copilot]\n  launch -- <args...>\n  --madmax [args...]                          (bare-flag launch with permissions bypass; alias of --yolo)\n  team <N:role> "<task>" [--name <name>] [--json]\n  team status <name> [--json]\n  team shutdown <name> [--json]\n  team api claim-task --input '<json>' [--json]\n  team api transition-task-status --input '<json>' [--json]\n  team api send-message --input '<json>' [--json]\n  team api broadcast --input '<json>' [--json]\n  team api mailbox-list --input '<json>' [--json]\n  team api mailbox-mark-delivered --input '<json>' [--json]\n  council "<question>" [--models a,b,c|m:role:weight] [--context <text|@file>] [--rubric <text|@file>] [--synth <model>] [--probe] [--timeout <ms>] [--synth-timeout <ms>] [--min-survivors <n>] [--max-concurrency <n>] [--tmp-dir <dir>] [--json]\n  comms status [--session <name>] [--json]      (is copilot on + online? auto-discovers session)\n  comms send --text "<prompt>" [--force] [--session <name>] [--json]\n  comms recv [--wait] [--lines <n>] [--timeout <ms>] [--session <name>] [--json]\n  comms ask --text "<prompt>" [--force] [--lines <n>] [--timeout <ms>] [--session <name>] [--json]\n  gateway serve [--only <name>[,<name>]]        (run all configured connectors; today: slack)\n  gateway status [--json] [--only <name>[,...]] (per-connector readiness; no sockets opened)\n  gateway doctor [--json] [--only <name>[,...]] (alias for 'gateway status')\n  gateway notify --text "<msg>" [--target slack:C\\|D\\|G\\|U... [:thread_ts]] [--thread-ts <ts>] [--json]\n                                                (one-shot outbound Slack post; falls back to SLACK_HOME_CHANNEL)\n  slack serve                                   (deprecated alias for 'gateway serve --only slack')\n  slack doctor [--json]                         (deprecated alias for 'gateway status --only slack')\n  env init [--force]                            (interactive: write ~/.omp/.env with Slack tokens + optional SLACK_HOME_CHANNEL)\n                                                non-interactive: set OMP_INIT_BOT_TOKEN/OMP_INIT_APP_TOKEN/OMP_INIT_HOME_CHANNEL\n                                                (env vars preferred over --bot-token/--app-token/--home-channel flags)\n  (--session is optional when exactly one omp-<digits> tmux session is running)\n${registeredCommandHelpLines().join("\n")}\n  ralph start "<task>" [--max-iterations <n>] [--session-id <id>] [--json]\n  ralph status [--json]\n  ralph tick [--json]\n  ralph cancel [--json]\n  ultrawork start "<objective>" [--task-count <n>] [--summary <s>] [--json]\n  ultrawork status [--json]\n  ultrawork cancel [--json]\n  ultraqa start "<goal>" [--max-cycles <n>] [--json]\n  ultraqa cycle pass|fail|pending [--json]\n  ultraqa status [--json]\n  ultraqa cancel [--json]\n  schedule add --id <id> --cron "<expr>" --prompt "<text>" [--bin copilot] [--model <m>] [--cwd <dir>] [--timeout <ms>] [--max-runs <n>] [--ttl-hours <h>] [--allow-all-tools] [--notify-target slack:<ID>] [--dry-run] [--json]\n  schedule list [--json]\n  schedule status <id> [--json]\n  schedule run-now <id> [--json]\n  schedule remove <id> [--json]\n  goal set "<objective>" [--json]\n  goal read [--json]\n  memory sync [--json]                          (render goal+directives into copilot-instructions.md)\n  daily-log set-goal "<text>" [--json]\n  daily-log add "<text>" [--json]\n  daily-log read [--days <n>] [--json]\n  daily-log prune [--keep-days <n>] [--json]\n  state write <key> <val> [--ttl <s>] | read|delete|status <key> | list | cleanup [--json]\n  project-memory read [<id>] | index | add-note "<title>" [--body "<text>"] | add-directive "<rule>" [--json]\n  trace timeline [<sessionId>] [--limit <n>] | summary [<sessionId>] | add <sessionId> <event> [<json>] [--json]\n  catalog list [--json]\n  catalog validate [--json]\n  catalog capability <id> [--json]\n  project inspect [--json]\n  skill install <skill-dir> [--root <repo>] [--scope project|user] [--dry-run] [--json]\n  lint:skills [--root <repo>]\n  sync:dry-run [--root <repo>]\n  jira:dry-run [--root <repo>]\n  jira render <plan-file> [--root <repo>] [--json]\n  jira apply <ticket-key-or-plan-file> --comment|--update|--transition|--link [--dry-run] [--json]\n`;
+    return `oh-my-copilot\n\nRun \`omp\` with no arguments to launch copilot (permissions bypass OFF).\nUse \`omp help\` to show this list.\n\nCommands:\n  (no args)                                     launch copilot (bypass OFF by default)\n  version [--json]\n  list [--json]\n  setup [--dry-run] [--scope project|user] [--plugin-root <dir>] [--json]\n  doctor [--json] [--copilot-bin <path>] [--skip-copilot] [--hooks]\n  cost [--json] [--session <id>] [--days <n>]\n  launch -- <args...>\n  --madmax [args...]                          (bare-flag launch with permissions bypass; alias of --yolo)\n  team <N:role> "<task>" [--name <name>] [--json]\n  team status <name> [--json]\n  team shutdown <name> [--json]\n  team api claim-task --input '<json>' [--json]\n  team api transition-task-status --input '<json>' [--json]\n  team api send-message --input '<json>' [--json]\n  team api broadcast --input '<json>' [--json]\n  team api mailbox-list --input '<json>' [--json]\n  team api mailbox-mark-delivered --input '<json>' [--json]\n  council "<question>" [--models a,b,c|m:role:weight] [--context <text|@file>] [--rubric <text|@file>] [--synth <model>] [--probe] [--timeout <ms>] [--synth-timeout <ms>] [--min-survivors <n>] [--max-concurrency <n>] [--tmp-dir <dir>] [--json]\n  comms status [--session <name>] [--json]      (is copilot on + online? auto-discovers session)\n  comms send --text "<prompt>" [--force] [--session <name>] [--json]\n  comms recv [--wait] [--lines <n>] [--timeout <ms>] [--session <name>] [--json]\n  comms ask --text "<prompt>" [--force] [--lines <n>] [--timeout <ms>] [--session <name>] [--json]\n  gateway serve [--only <name>[,<name>]]        (run all configured connectors; today: slack)\n  gateway status [--json] [--only <name>[,...]] (per-connector readiness; no sockets opened)\n  gateway doctor [--json] [--only <name>[,...]] (alias for 'gateway status')\n  gateway notify --text "<msg>" [--target slack:C\\|D\\|G\\|U... [:thread_ts]] [--thread-ts <ts>] [--json]\n                                                (one-shot outbound Slack post; falls back to SLACK_HOME_CHANNEL)\n  slack serve                                   (deprecated alias for 'gateway serve --only slack')\n  slack doctor [--json]                         (deprecated alias for 'gateway status --only slack')\n  env init [--force]                            (interactive: write ~/.omp/.env with Slack tokens + optional SLACK_HOME_CHANNEL)\n                                                non-interactive: set OMP_INIT_BOT_TOKEN/OMP_INIT_APP_TOKEN/OMP_INIT_HOME_CHANNEL\n                                                (env vars preferred over --bot-token/--app-token/--home-channel flags)\n  (--session is optional when exactly one omp-<digits> tmux session is running)\n${registeredCommandHelpLines().join("\n")}\n  ralph start "<task>" [--max-iterations <n>] [--session-id <id>] [--json]\n  ralph status [--json]\n  ralph tick [--json]\n  ralph cancel [--json]\n  ultrawork start "<objective>" [--task-count <n>] [--summary <s>] [--json]\n  ultrawork status [--json]\n  ultrawork cancel [--json]\n  ultraqa start "<goal>" [--max-cycles <n>] [--json]\n  ultraqa cycle pass|fail|pending [--json]\n  ultraqa status [--json]\n  ultraqa cancel [--json]\n  schedule add --id <id> --cron "<expr>" --prompt "<text>" [--bin copilot] [--model <m>] [--cwd <dir>] [--timeout <ms>] [--max-runs <n>] [--ttl-hours <h>] [--allow-all-tools] [--notify-target slack:<ID>] [--dry-run] [--json]\n  schedule list [--json]\n  schedule status <id> [--json]\n  schedule run-now <id> [--json]\n  schedule remove <id> [--json]\n  goal set "<objective>" [--json]\n  goal read [--json]\n  memory sync [--json]                          (render goal+directives into copilot-instructions.md)\n  config get [--json] | config set memory-mode on|off | config set memory-review-model <slug> | config set memory-review-min-messages <n> [--global]\n                                                (--global writes ~/.omp/config.json; applies to every project. project .omp/config.json overrides it)\n  memory-review --session <uuid|latest> [--model <slug>] [--json]   (cheap-model end-of-session review; opt-in via memory-mode)\n  daily-log set-goal "<text>" [--json]\n  daily-log add "<text>" [--json]\n  daily-log read [--days <n>] [--json]\n  daily-log prune [--keep-days <n>] [--json]\n  state write <key> <val> [--ttl <s>] | read|delete|status <key> | list | cleanup [--json]\n  project-memory read [<id>] | index | add-note "<title>" [--body "<text>"] | add-directive "<rule>" | prune-notes --keep <n>|--older-than <days> [--json]\n  trace timeline [<sessionId>] [--limit <n>] | summary [<sessionId>] | add <sessionId> <event> [<json>] [--json]\n  catalog list [--json]\n  catalog validate [--json]\n  catalog capability <id> [--json]\n  project inspect [--json]\n  skill install <skill-dir> [--root <repo>] [--scope project|user] [--dry-run] [--json]\n  lint:skills [--root <repo>]\n  sync:dry-run [--root <repo>]\n  jira:dry-run [--root <repo>]\n  jira render <plan-file> [--root <repo>] [--json]\n  jira apply <ticket-key-or-plan-file> --comment|--update|--transition|--link [--dry-run] [--json]\n`;
 }
 async function resolveExistingInputPath(value) {
     const { existsSync } = await import("node:fs");
@@ -40,6 +40,33 @@ async function resolveExistingInputPath(value) {
     return direct;
 }
 const BARE_LAUNCH_FLAGS = new Set(["--madmax", "--yolo"]);
+// Snapshot existing Copilot session dirs BEFORE launch, so after a headless run
+// we can identify the exact session it created (instead of guessing "latest").
+async function snapshotSessionsForReview() {
+    try {
+        const { listSessionIds } = await import("./memory-review/transcript.js");
+        return listSessionIds();
+    }
+    catch {
+        return [];
+    }
+}
+// Headless `copilot -p` skips hooks, so when omp launched copilot in headless
+// mode we detach the end-of-session memory review here. We review the session
+// that appeared since `before` — never a guessed "latest" — and skip if none
+// is identifiable. Best-effort: a failure here must never affect the launch.
+async function maybeTriggerHeadlessReview(argv, cwd, before) {
+    try {
+        const { triggerHeadlessReview } = await import("./memory-review/trigger.js");
+        const { newestSessionSince } = await import("./memory-review/transcript.js");
+        const cliPath = join(packageRootFromImportMeta(import.meta.url), "dist", "src", "cli.js");
+        const sessionId = newestSessionSince(before) ?? "";
+        triggerHeadlessReview({ cwd, argv, cliPath, sessionId });
+    }
+    catch {
+        // never fail a launch on the review trigger
+    }
+}
 export async function runCli(argv = process.argv.slice(2)) {
     const [group, command, value] = argv;
     const json = hasFlag(argv, "--json");
@@ -70,11 +97,14 @@ export async function runCli(argv = process.argv.slice(2)) {
     // case argv is empty, so normalizeCopilotLaunchArgs emits no --yolo.
     if (!group || BARE_LAUNCH_FLAGS.has(group)) {
         const { launchCopilot } = await import("./copilot/launch.js");
+        const launchCwd = flagValue(argv, "--root") ?? process.cwd();
+        const beforeSessions = await snapshotSessionsForReview();
         const result = await launchCopilot({
             args: argv,
             bin: flagValue(argv, "--bin"),
-            cwd: flagValue(argv, "--root") ?? process.cwd(),
+            cwd: launchCwd,
         });
+        await maybeTriggerHeadlessReview(argv, launchCwd, beforeSessions);
         return json
             ? { ok: result.ok, exitCode: result.exitCode, output: result }
             : {
@@ -107,20 +137,32 @@ export async function runCli(argv = process.argv.slice(2)) {
             importMetaUrl: import.meta.url,
             copilotBin: flagValue(argv, "--copilot-bin"),
             skipCopilot: hasFlag(argv, "--skip-copilot"),
+            checkHooks: hasFlag(argv, "--hooks"),
         });
         return json
             ? { ok: report.ok, exitCode: report.ok ? 0 : 1, output: report }
             : { ok: report.ok, exitCode: report.ok ? 0 : 1, message: formatDoctor(report) };
     }
+    if (group === "cost") {
+        const { summarizeCost, formatCostSummary } = await import("./cost/summary.js");
+        const summary = summarizeCost(flagValue(argv, "--root") ?? process.cwd(), {
+            sessionId: flagValue(argv, "--session"),
+            today: hasFlag(argv, "--today"),
+        });
+        return json ? { ok: true, output: summary } : { ok: true, message: formatCostSummary(summary) };
+    }
     if (group === "launch") {
         const dashIndex = argv.indexOf("--");
         const passthrough = dashIndex >= 0 ? argv.slice(dashIndex + 1) : argv.slice(1);
         const { launchCopilot } = await import("./copilot/launch.js");
+        const launchCwd = flagValue(argv, "--root") ?? process.cwd();
+        const beforeSessions = await snapshotSessionsForReview();
         const result = await launchCopilot({
             args: passthrough,
             bin: flagValue(argv, "--bin"),
-            cwd: flagValue(argv, "--root") ?? process.cwd(),
+            cwd: launchCwd,
         });
+        await maybeTriggerHeadlessReview(passthrough, launchCwd, beforeSessions);
         return json
             ? { ok: result.ok, exitCode: result.exitCode, output: result }
             : {
@@ -192,6 +234,80 @@ export async function runCli(argv = process.argv.slice(2)) {
         }
         return { ok: false, exitCode: 1, message: "Unknown memory subcommand. Try: memory sync" };
     }
+    if (group === "config") {
+        const { readMemoryConfig, setMemoryConfigValue } = await import("./memory-review/config.js");
+        const cwd = flagValue(argv, "--root") ?? process.cwd();
+        // OMP_HOME_OVERRIDE relocates the global ~/.omp config dir (test seam; also
+        // lets users point at a custom home). Undefined => os.homedir() default.
+        const homeDir = process.env.OMP_HOME_OVERRIDE || undefined;
+        // `--global` writes to ~/.omp/config.json so the setting applies everywhere.
+        const scope = hasFlag(argv, "--global") ? "global" : "project";
+        if (command === "get" || command === undefined) {
+            const cfg = readMemoryConfig(cwd, { homeDir });
+            return json
+                ? { ok: true, output: cfg }
+                : {
+                    ok: true,
+                    message: `memory-mode=${cfg.memoryMode}\nmemory-review-model=${cfg.memoryReviewModel}\nmemory-review-min-messages=${cfg.memoryReviewMinMessages}`,
+                };
+        }
+        if (command === "set") {
+            const setVal = argv[3];
+            const where = scope === "global" ? " (global ~/.omp)" : "";
+            if (value === "memory-mode") {
+                if (setVal !== "on" && setVal !== "off") {
+                    return { ok: false, exitCode: 1, message: "usage: omp config set memory-mode on|off [--global]" };
+                }
+                setMemoryConfigValue(cwd, "memoryMode", setVal, { scope, homeDir });
+                return json ? { ok: true, output: { memoryMode: setVal, scope } } : { ok: true, message: `memory-mode=${setVal}${where}` };
+            }
+            if (value === "memory-review-model") {
+                if (!setVal || setVal.startsWith("-")) {
+                    return { ok: false, exitCode: 1, message: "usage: omp config set memory-review-model <slug> [--global]" };
+                }
+                setMemoryConfigValue(cwd, "memoryReviewModel", setVal, { scope, homeDir });
+                return json ? { ok: true, output: { memoryReviewModel: setVal, scope } } : { ok: true, message: `memory-review-model=${setVal}${where}` };
+            }
+            if (value === "memory-review-min-messages") {
+                const n = Number(setVal);
+                if (!Number.isFinite(n) || n < 0) {
+                    return { ok: false, exitCode: 1, message: "usage: omp config set memory-review-min-messages <non-negative integer> [--global]" };
+                }
+                setMemoryConfigValue(cwd, "memoryReviewMinMessages", String(Math.floor(n)), { scope, homeDir });
+                return json ? { ok: true, output: { memoryReviewMinMessages: Math.floor(n), scope } } : { ok: true, message: `memory-review-min-messages=${Math.floor(n)}${where}` };
+            }
+            return { ok: false, exitCode: 1, message: "Unknown config key. Try: memory-mode | memory-review-model | memory-review-min-messages" };
+        }
+        return { ok: false, exitCode: 1, message: "Unknown config subcommand. Try: config get | config set <key> <value>" };
+    }
+    if (group === "memory-review") {
+        const { runMemoryReview } = await import("./memory-review/index.js");
+        const { createReviewSpawn } = await import("./memory-review/spawn.js");
+        const { isValidSessionId, latestSessionId } = await import("./memory-review/transcript.js");
+        const cwd = flagValue(argv, "--root") ?? process.cwd();
+        let sessionId = flagValue(argv, "--session") ?? "";
+        if (!sessionId || sessionId === "latest") {
+            sessionId = latestSessionId() ?? "";
+        }
+        if (!sessionId) {
+            return { ok: false, exitCode: 1, message: "usage: omp memory-review --session <uuid|latest>" };
+        }
+        if (!isValidSessionId(sessionId)) {
+            return { ok: false, exitCode: 1, message: "invalid --session id" };
+        }
+        const res = await runMemoryReview({
+            cwd,
+            sessionId,
+            spawn: createReviewSpawn(flagValue(argv, "--bin")),
+            model: flagValue(argv, "--model"),
+        });
+        return json
+            ? { ok: true, output: res }
+            : {
+                ok: true,
+                message: res.ran ? `memory-review ran: ${JSON.stringify(res.summary)}` : `memory-review skipped: ${res.reason}`,
+            };
+    }
     if (group === "daily-log") {
         const { setDailyGoal, addLogEntry, readDailyLog, pruneDailyLog } = await import("./daily-log.js");
         const cwd = flagValue(argv, "--root") ?? process.cwd();
@@ -304,6 +420,20 @@ export async function runCli(argv = process.argv.slice(2)) {
         if (command === "index") {
             return { ok: true, output: { notes: pm.noteIndex(cwd) } };
         }
+        if (command === "prune-notes") {
+            const keepRaw = flagValue(argv, "--keep");
+            const olderRaw = flagValue(argv, "--older-than");
+            const keep = keepRaw !== undefined && Number.isFinite(Number(keepRaw)) ? Number(keepRaw) : undefined;
+            const olderThanDays = olderRaw !== undefined && Number.isFinite(Number(olderRaw)) ? Number(olderRaw) : undefined;
+            if (keep === undefined && olderThanDays === undefined) {
+                return { ok: false, exitCode: 1, message: "usage: omp project-memory prune-notes --keep <n> | --older-than <days>" };
+            }
+            const removed = pm.pruneNotes(cwd, { keep, olderThanDays });
+            syncInstructionsMemory(cwd); // refresh the managed block after pruning
+            return json
+                ? { ok: true, output: { removed } }
+                : { ok: true, message: `pruned ${removed.length} note${removed.length === 1 ? "" : "s"}` };
+        }
         if (command === "read" || command === undefined) {
             // `read <id>` loads one note's body on demand; bare `read` returns the
             // bounded summary (directives + note index — never note bodies).
@@ -318,7 +448,7 @@ export async function runCli(argv = process.argv.slice(2)) {
         return {
             ok: false,
             exitCode: 1,
-            message: 'Unknown project-memory subcommand. Try: project-memory read [<id>] | index | add-note "<title>" [--body "<text>"] | add-directive "<rule>"',
+            message: 'Unknown project-memory subcommand. Try: project-memory read [<id>] | index | add-note "<title>" [--body "<text>"] | add-directive "<rule>" | prune-notes --keep <n>|--older-than <days>',
         };
     }
     if (group === "trace") {