drip-cli 0.1.1

Delta Read Interception Proxy — sends only file diffs to your LLM agent
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
use anyhow::Result;
use clap::{Parser, Subcommand, ValueEnum};

mod commands;
mod core;
mod hooks;
mod mcp;

#[derive(Parser, Debug)]
#[command(
    name = "drip",
    version,
    about = "Delta Read Interception Proxy — send only diffs to LLM agents.",
    propagate_version = true
)]
pub struct Cli {
    #[command(subcommand)]
    command: Cmd,
}

#[derive(Subcommand, Debug)]
pub enum Cmd {
    /// Read a file: full content on first read, unified diff afterwards.
    Read {
        file: String,
        /// Compute the outcome but do not update the session baseline.
        /// Useful for previewing what DRIP would send to the agent.
        #[arg(long)]
        dry_run: bool,
    },
    /// Install hooks for a supported agent.
    Init {
        /// Install at the user level (~/.claude/) instead of project-level.
        #[arg(short = 'g', long)]
        global: bool,
        /// Target agent: `claude`, `codex`, or `gemini`.
        #[arg(long, default_value = "claude")]
        agent: String,
    },
    /// Reverse of `init`: prune hooks, remove the @drip.md include
    /// from CLAUDE.md, delete drip.md. Pre-existing user content in
    /// CLAUDE.md is preserved.
    Uninstall {
        /// Uninstall from ~/.claude/ instead of the project tree.
        #[arg(short = 'g', long)]
        global: bool,
        /// Target agent: `claude`, `codex`, or `gemini`.
        #[arg(long, default_value = "claude")]
        agent: String,
    },
    /// Show token-savings statistics. Defaults to *cumulative since
    /// install*; pass --session to scope to the current session only.
    Meter {
        /// Include a per-day breakdown for the last 30 days.
        #[arg(long)]
        history: bool,
        /// Render an ASCII bar of sent vs full tokens.
        #[arg(long)]
        graph: bool,
        /// Emit machine-readable JSON instead of the human report.
        #[arg(long)]
        json: bool,
        /// Scope the report to a specific session. Bare `--session` (no
        /// value) means the current shell session — useful right after
        /// `drip reset` or to verify the agent's session in isolation.
        /// `--session <id>` targets that exact id (run `drip sessions`
        /// to list them). Without the flag, `drip meter` aggregates every
        /// read since DRIP was first installed.
        #[arg(long, num_args = 0..=1, default_missing_value = "")]
        session: Option<String>,
        /// Drop entries for files that no longer exist on disk before
        /// printing the report. Useful when benches or `/tmp` files have
        /// inflated the lifetime counters with artifacts. The headline
        /// ratio gets recomputed from the survivors.
        #[arg(long)]
        prune: bool,
    },
    /// Clear DRIP state. Default: just the current session's tracked
    /// reads (cheap, reversible — the next read becomes a fresh first
    /// read). Pass `--all` to nuke EVERY session, lifetime counter,
    /// and on-disk blob — that one is irreversible.
    /// `--stats` zeros the lifetime counters only, leaving baselines
    /// intact.
    Reset {
        /// Delete ALL DRIP data — every session, baseline,
        /// file-registry entry, blob on disk, and the lifetime
        /// counters. Use to start fresh after a bench run pollutes
        /// counters or to wipe the box before uninstall. Asks for
        /// `yes` confirmation unless `--force`.
        #[arg(long)]
        all: bool,
        /// Zero `lifetime_stats` / `lifetime_per_file` / `lifetime_daily`
        /// / `lifetime_edited_files`. Leaves
        /// active sessions and per-file baselines untouched, so an
        /// in-progress agent run keeps its diffs / sentinels.
        #[arg(long)]
        stats: bool,
        /// Skip the interactive confirmation prompt for `--all`.
        /// Required in scripts and CI; ignored otherwise.
        #[arg(long)]
        force: bool,
    },
    /// Drop the cached baseline for one file so the next read returns
    /// full content. Use after an out-of-band edit (manual change in
    /// another editor, `git pull`, …).
    Refresh { file: String },
    /// List recent sessions stored in the SQLite database. A bare
    /// `list` / `ls` positional is accepted as a no-op alias for
    /// symmetry with multi-action CLIs (e.g. `drip cache stats`).
    Sessions {
        /// Optional alias: `list` or `ls`. Both render the same table
        /// as plain `drip sessions`.
        #[arg(value_parser = ["list", "ls"], hide = true)]
        _alias: Option<String>,
    },
    /// Hook entry point — reads a JSON payload from stdin, writes the
    /// rewritten response to stdout. Used by the agent integrations.
    Hook {
        #[arg(value_enum)]
        agent: HookAgentArg,
    },
    /// Run as an MCP (Model Context Protocol) stdio server.
    /// Used by Codex CLI and other MCP-compatible agents.
    Mcp {
        /// Self-identify the calling agent so `drip sessions` can
        /// distinguish Codex / Gemini from a regular shell.
        /// `drip init --agent <X>` writes this flag into the MCP
        /// config it generates; users invoking `drip mcp` directly
        /// can pass it themselves.
        #[arg(long)]
        agent: Option<String>,
    },
    /// Run a background watcher that pre-computes diffs for tracked
    /// files. The Read hook then serves the precomputed result instead
    /// of running the diff inline — making the hook effectively
    /// instantaneous on big files.
    Watch {
        /// Directory to watch (defaults to the current working directory).
        path: Option<std::path::PathBuf>,
    },
    /// Manage the file-cache directory used for large `reads.content`
    /// blobs. Subcommands: `gc` removes orphan blobs whose hash no row
    /// references; `stats` shows inline-vs-file row counts, dedup
    /// hits, on-disk size.
    Cache {
        #[command(subcommand)]
        action: CacheAction,
    },
    /// Manage the cross-session file registry. Subcommands: `stats`
    /// (counts of known files, oldest/most-accessed); `gc` (drop
    /// entries unseen for `--older-than DURATION`, default 30d).
    Registry {
        #[command(subcommand)]
        action: RegistryAction,
    },
    /// Generate a shell completion script on stdout. Pipe it into the
    /// conventional file for your shell:
    ///   drip completions zsh  > ~/.zsh/completions/_drip
    ///   drip completions bash > ~/.bash_completion.d/drip.bash
    ///   drip completions fish > ~/.config/fish/completions/drip.fish
    /// `drip init` does this automatically for the detected `$SHELL`.
    Completions {
        /// Shell to generate completions for: bash, zsh, or fish.
        shell: String,
    },
    /// Diagnose the install — binary, DB, file cache, agent hooks,
    /// drip.md/CLAUDE.md, shell completions, current session. Exit
    /// code 0 if no errors, 1 otherwise.
    Doctor {
        /// Emit machine-readable JSON instead of the human report.
        #[arg(long)]
        json: bool,
        /// Print only the final summary line (exit code remains
        /// significant for scripting).
        #[arg(long)]
        quiet: bool,
    },
    /// Upgrade DRIP to the latest released version. Detects how
    /// DRIP was installed (Homebrew, `cargo install`, install
    /// script) and runs the matching upgrade command. No-op when
    /// already on the latest tag.
    Update {
        /// Show what would be done without actually invoking
        /// brew / cargo / the install script.
        #[arg(long)]
        dry_run: bool,
    },
    /// Inspect the compressed→original line map persisted for a
    /// semantically-compressed read. Without --line, prints the full
    /// table (one row per compressed line). With --line N, resolves
    /// just that line — useful when an Edit landed on `L5` of the
    /// stubbed view and you need the real source range to reason about.
    SourceMap {
        /// Path to the file. The file must have been read in the
        /// current session (`drip read <file>` or via a hook) AND
        /// compression must have fired for a map to exist.
        file: String,
        /// Resolve a single compressed line. Accepts `5` or `L5`.
        #[arg(long)]
        line: Option<String>,
        /// Emit JSON (mirrors the persisted column shape).
        #[arg(long)]
        json: bool,
    },
    /// Replay an agent session — show every read, in order, with the
    /// exact bytes DRIP returned. Killer for debugging "why did the
    /// agent see X here?" without re-running the workflow.
    Replay {
        /// Session id to replay (default: most recent active with events).
        #[arg(long)]
        session: Option<String>,
        /// Filter to events from the last N units (e.g. 5m, 1h, 2d).
        #[arg(long)]
        since: Option<String>,
        /// Filter to events whose path contains this substring.
        #[arg(long)]
        file: Option<String>,
        /// Max number of events to display (default 50).
        #[arg(long)]
        limit: Option<i64>,
        /// Print the full rendered output for each event, not just the
        /// summary table.
        #[arg(long)]
        full: bool,
        /// Emit machine-readable JSON instead of the human report.
        #[arg(long)]
        json: bool,
    },
}

#[derive(Subcommand, Debug)]
pub enum RegistryAction {
    /// Print known-file count, inline/cache breakdown, oldest entry,
    /// most-accessed file.
    Stats,
    /// Drop registry entries whose `last_seen_at` is older than
    /// `--older-than` (default `30d`). Cascade-deletes orphaned
    /// cache blobs.
    Gc {
        /// Age threshold: `30d`, `12h`, `90m`, `45s`. Bare digits
        /// mean seconds. Default: 30d.
        #[arg(long, default_value = "30d")]
        older_than: String,
    },
}

#[derive(Subcommand, Debug)]
pub enum CacheAction {
    /// Remove cache blobs whose hash isn't referenced by any `reads`
    /// row (typically: rows expired by the 2 h session purge).
    Gc,
    /// Print inline / file-cache row counts, dedup hits, sizes.
    Stats,
    /// Hoist any inline `reads.content` rows whose payload is larger
    /// than `DRIP_INLINE_MAX_BYTES` into the file cache, then VACUUM
    /// the SQLite file to actually reclaim disk. Useful right after
    /// upgrading from v1, or when a benchmark has bloated the DB
    /// with huge inline payloads.
    Compact,
}

#[derive(ValueEnum, Clone, Debug)]
pub enum HookAgentArg {
    Claude,
    ClaudeGlob,
    ClaudeGrep,
    ClaudePostEdit,
    ClaudePreEdit,
    ClaudeSessionStart,
    Gemini,
    GeminiCompress,
}

impl From<HookAgentArg> for commands::hook::HookAgent {
    fn from(v: HookAgentArg) -> Self {
        match v {
            HookAgentArg::Claude => commands::hook::HookAgent::Claude,
            HookAgentArg::ClaudeGlob => commands::hook::HookAgent::ClaudeGlob,
            HookAgentArg::ClaudeGrep => commands::hook::HookAgent::ClaudeGrep,
            HookAgentArg::ClaudePostEdit => commands::hook::HookAgent::ClaudePostEdit,
            HookAgentArg::ClaudePreEdit => commands::hook::HookAgent::ClaudePreEdit,
            HookAgentArg::ClaudeSessionStart => commands::hook::HookAgent::ClaudeSessionStart,
            HookAgentArg::Gemini => commands::hook::HookAgent::Gemini,
            HookAgentArg::GeminiCompress => commands::hook::HookAgent::GeminiCompress,
        }
    }
}

fn main() -> Result<()> {
    // Hook-tolerance shim: if argv looks like `drip hook <unknown>`,
    // emit a no-op JSON response and exit 0 instead of letting clap
    // crash. This is the upgrade-path safety net — a stale
    // `settings.json` after a `drip` binary upgrade (e.g. a removed
    // subcommand like the old `claude-bash`) used to take down every
    // matching agent tool call until the user re-ran `drip init`.
    // Now the agent just sees its tool fire natively; DRIP records
    // nothing for the call but doesn't crash the integration.
    let cli = match Cli::try_parse() {
        Ok(cli) => cli,
        Err(e) => {
            return handle_clap_error(e);
        }
    };

    let out = match cli.command {
        Cmd::Read { file, dry_run } => commands::read::run_with(&file, dry_run)?,
        Cmd::Init { global, agent } => {
            let agent = commands::init::Agent::parse(&agent)?;
            commands::init::run(agent, global)?
        }
        Cmd::Uninstall { global, agent } => {
            let agent = commands::init::Agent::parse(&agent)?;
            commands::init::run_uninstall(agent, global)?
        }
        Cmd::Meter {
            history,
            graph,
            json,
            session,
            prune,
        } => {
            // `Some("")` = bare `--session`, `Some(id)` = explicit target,
            // `None` = no flag (default lifetime view).
            let (session_only, session_id) = match session {
                None => (false, None),
                Some(s) if s.is_empty() => (true, None),
                Some(id) => (true, Some(id)),
            };
            commands::meter::run(commands::meter::MeterOpts {
                history,
                graph,
                json,
                session_only,
                session_id,
                prune,
            })?
        }
        Cmd::Reset { all, stats, force } => {
            commands::reset::run(commands::reset::ResetOpts { all, stats, force })?
        }
        Cmd::Refresh { file } => commands::refresh::run(&file)?,
        Cmd::Sessions { _alias: _ } => commands::sessions::run()?,
        Cmd::Hook { agent } => commands::hook::run(agent.into())?,
        Cmd::Mcp { agent } => {
            // Tag BEFORE Session::open so $DRIP_AGENT is set in time.
            // Don't clobber a manually-exported value when --agent
            // isn't passed.
            if let Some(a) = agent {
                std::env::set_var("DRIP_AGENT", a);
            }
            mcp::run()?;
            return Ok(());
        }
        Cmd::Watch { path } => commands::watch::run(path)?,
        Cmd::Cache { action } => match action {
            CacheAction::Gc => commands::cache::run_gc()?,
            CacheAction::Stats => commands::cache::run_stats()?,
            CacheAction::Compact => commands::cache::run_compact()?,
        },
        Cmd::Registry { action } => match action {
            RegistryAction::Stats => commands::registry::run_stats()?,
            RegistryAction::Gc { older_than } => {
                let secs = commands::registry::parse_duration(&older_than);
                commands::registry::run_gc(secs)?
            }
        },
        Cmd::Update { dry_run } => {
            commands::update::run(dry_run)?;
            return Ok(());
        }
        Cmd::SourceMap { file, line, json } => {
            let line = match line {
                None => None,
                Some(raw) => Some(commands::source_map::parse_line_arg(&raw)?),
            };
            commands::source_map::run(commands::source_map::Opts { file, line, json })?
        }
        Cmd::Replay {
            session,
            since,
            file,
            limit,
            full,
            json,
        } => commands::replay::run(commands::replay::ReplayOpts {
            session,
            since,
            file,
            limit,
            full,
            json,
        })?,
        Cmd::Completions { shell } => {
            commands::completions::run(&shell)?;
            return Ok(());
        }
        Cmd::Doctor { json, quiet } => {
            let (out, code) = commands::doctor::run(commands::doctor::DoctorOpts { json, quiet })?;
            print!("{out}");
            if !out.ends_with('\n') {
                println!();
            }
            std::process::exit(code);
        }
    };

    print!("{out}");
    if !out.ends_with('\n') {
        println!();
    }
    Ok(())
}

/// Bridge between clap's strict subcommand parsing and DRIP's
/// "never crash a tool call" invariant. Most parse errors are honest
/// user mistakes (`drip met` instead of `drip meter`) and should hit
/// clap's normal help/usage exit. The exception is `drip hook
/// <invalid>` — argv[1]=="hook", argv[2] doesn't match any current
/// subcommand. That's the upgrade-path footgun: a stale settings.json
/// from a binary that knew `claude-bash` (or any future-removed
/// hook), pointing at a newer `drip` that doesn't. Without this
/// shim, the hook crashes, Claude Code reports "blocking error", and
/// every matching tool call fails until the user re-runs init. With
/// it, we drain the stdin payload (some hooks send big payloads;
/// not draining could deadlock on PIPE) and emit `{}` — Claude Code
/// treats an empty object as "no opinion, proceed normally", which
/// degrades the no-longer-existing hook into a silent passthrough.
fn handle_clap_error(err: clap::Error) -> Result<()> {
    let args: Vec<String> = std::env::args().collect();
    let is_stale_hook = args.len() >= 3
        && args[1] == "hook"
        && matches!(
            err.kind(),
            clap::error::ErrorKind::InvalidValue
                | clap::error::ErrorKind::UnknownArgument
                | clap::error::ErrorKind::InvalidSubcommand
        );
    if is_stale_hook {
        // Drain stdin so the hook caller's write doesn't block. The
        // 32 KB cap matches the rolling event log; payloads bigger
        // than that get truncated, but Claude Code's hook payloads
        // are well under that.
        use std::io::Read;
        let mut sink = Vec::with_capacity(4096);
        let _ = std::io::stdin().lock().take(32_768).read_to_end(&mut sink);
        // One-shot stderr breadcrumb so a curious user running with
        // hook-debug logs can see what happened. Claude Code ignores
        // stderr from PreToolUse hooks.
        eprintln!(
            "drip: ignoring stale hook subcommand `{}` — re-run `drip init` to clean up settings.json",
            args.get(2).map(String::as_str).unwrap_or("?")
        );
        println!("{{}}");
        return Ok(());
    }
    err.exit();
}