devflow_core/agents/
mod.rs

1//! Agent adapter trait and implementations.
2//!
3//! Each adapter knows how to wrap a stage prompt into its CLI's non-interactive
4//! launch command. The prompt text itself comes from [`crate::prompt`] — the
5//! adapter only formats it into the right flags for its agent.
6
7use crate::state::AgentKind;
8use std::path::PathBuf;
9
10/// Common behavior implemented by every supported coding-agent backend.
11pub trait AgentAdapter {
12    /// Human-readable adapter name.
13    fn name(&self) -> &'static str;
14
15    /// Build the command and arguments to launch this agent headless with the
16    /// given `prompt` for `phase`. Returns `(program, args)`.
17    ///
18    /// `extra_writable_roots` are directories OUTSIDE the agent's working
19    /// directory that its sandbox must still be allowed to write. Linked git
20    /// worktrees keep their git metadata under the main repo's `.git/` — and
21    /// Codex additionally read-only-mounts the cwd's resolved git dir, so
22    /// BOTH the common `.git` and the worktree admin dir
23    /// (`.git/worktrees/<name>`) must be granted explicitly (13-06 dogfood
24    /// finding, verified with `codex sandbox` probes). Adapters without a
25    /// sandbox ignore it.
26    fn exec_command(
27        &self,
28        phase: u32,
29        prompt: &str,
30        extra_writable_roots: &[PathBuf],
31    ) -> (&'static str, Vec<String>);
32
33    /// Extra environment variables for the agent process tree. Codex uses
34    /// this to disable commit/tag signing inside its sandbox: the operator's
35    /// signing agent (ssh-agent/gpg-agent) is unreachable there, so signed
36    /// commits fail headless with a passphrase error (13-06 dogfood finding
37    /// — same rationale as the unsigned VersionBump tags). `GIT_CONFIG_*`
38    /// env scoping keeps the override out of every repo/global config.
39    fn extra_env(&self) -> Vec<(String, String)> {
40        Vec::new()
41    }
42
43    /// Detect an agent-specific completion signal in captured output.
44    fn completion_signal_detected(&self, output: &str) -> bool;
45
46    /// Adapter-specific pre-launch readiness check (D-13/D-14 adapter hook,
47    /// Phase 17c). The default is a no-op — most adapters have nothing extra
48    /// to check, mirroring [`Self::extra_env`]'s empty-default shape. The
49    /// `Err` variant is a human-readable failure reason that flows into the
50    /// preflight gate's context (`run_preflight` in `devflow-cli/src/main.rs`).
51    /// This is the trait surface Phase 18's Hermes adapter implements to
52    /// enforce a non-empty reviewer/receiver set — no built-in adapter
53    /// (Claude/Codex/OpenCode) overrides it in Phase 17 because no
54    /// reviewer-set storage exists yet in `state.rs`/`config.rs` (review
55    /// consensus #6).
56    fn preflight(&self, _state: &crate::state::State) -> Result<(), String> {
57        Ok(())
58    }
59}
60
61/// Return an adapter for a configured agent kind.
62pub fn adapter_for(kind: AgentKind) -> Box<dyn AgentAdapter> {
63    match kind {
64        AgentKind::Claude => Box::new(ClaudeAgent),
65        AgentKind::Codex => Box::new(CodexAgent),
66        AgentKind::OpenCode => Box::new(OpenCodeAgent),
67    }
68}
69
70pub mod claude;
71pub mod codex;
72pub mod opencode;
73
74pub use claude::ClaudeAgent;
75pub use codex::CodexAgent;
76pub use opencode::OpenCodeAgent;
77
78#[cfg(test)]
79mod tests {
80    use super::*;
81    use crate::prompt::stage_prompt;
82    use crate::stage::Stage;
83
84    #[test]
85    fn adapter_for_returns_correct_names() {
86        assert_eq!(adapter_for(AgentKind::Claude).name(), "Claude Code");
87        assert_eq!(adapter_for(AgentKind::Codex).name(), "OpenAI Codex");
88        assert_eq!(adapter_for(AgentKind::OpenCode).name(), "OpenCode");
89    }
90
91    /// Extract the prompt text as this adapter actually DELIVERS it.
92    ///
93    /// Codex and OpenCode pass it positionally, so it is read back out of
94    /// argv. Claude does not: under `--input-format stream-json` the initial
95    /// user turn travels on the child's stdin, so it is read back out of the
96    /// wire document [`crate::monitor::user_turn_line`] builds. Two lookups,
97    /// one question — "what text did the agent receive?".
98    fn delivered_prompt(kind: AgentKind, prompt: &str) -> String {
99        if kind == AgentKind::Claude {
100            let turn: serde_json::Value =
101                serde_json::from_str(&crate::monitor::user_turn_line(prompt))
102                    .expect("the stdin user turn must be one valid JSON document");
103            return turn
104                .get("message")
105                .and_then(|message| message.get("content"))
106                .and_then(serde_json::Value::as_str)
107                .expect("the user turn must carry the prompt as message.content")
108                .to_string();
109        }
110        let (_program, args) = adapter_for(kind).exec_command(7, prompt, &[]);
111        args.into_iter()
112            .find(|arg| arg.contains("DEVFLOW_RESULT"))
113            .expect("agent command should carry the prompt with the DEVFLOW_RESULT contract")
114    }
115
116    /// The invariant survived a transport change; it was not deleted with the
117    /// mechanism that used to carry it. Every adapter still receives the
118    /// canonical stage prompt byte-for-byte — Codex and OpenCode in argv,
119    /// Claude in the stdin user turn.
120    ///
121    /// The Claude leg additionally asserts the prompt is ABSENT from argv,
122    /// because "identical text" would otherwise be satisfiable by an adapter
123    /// that sent the prompt through both routes — which would double the
124    /// initial turn.
125    #[test]
126    fn every_adapter_receives_identical_prompt_text() {
127        let prompt = stage_prompt(Stage::Code, 7);
128        for kind in [AgentKind::Claude, AgentKind::Codex, AgentKind::OpenCode] {
129            assert_eq!(
130                delivered_prompt(kind, &prompt),
131                prompt,
132                "{kind} must receive the canonical stage prompt unchanged"
133            );
134        }
135
136        let (_program, args) = adapter_for(AgentKind::Claude).exec_command(7, &prompt, &[]);
137        assert!(
138            !args.iter().any(|arg| arg.contains("DEVFLOW_RESULT")),
139            "Claude's prompt must travel on stdin ONLY; a copy left in argv \
140             would deliver the initial turn twice: {args:?}"
141        );
142    }
143
144    /// The Phase 31 launch contract, asserted as one thing because getting
145    /// only the flags right is the documented way to half-implement it: the
146    /// transport is `stream-json` in BOTH directions, and the prompt is not a
147    /// positional argument at all.
148    #[test]
149    fn claude_launches_headless_stream_json_without_positional_prompt() {
150        let prompt = stage_prompt(Stage::Code, 3);
151        let (program, args) = adapter_for(AgentKind::Claude).exec_command(3, &prompt, &[]);
152        assert_eq!(program, "claude");
153        assert!(args.iter().any(|a| a == "-p"));
154        assert!(
155            args.windows(2)
156                .any(|w| w[0] == "--input-format" && w[1] == "stream-json"),
157            "the INPUT format is what moves the initial turn onto stdin; \
158             flipping only the output format leaves the CLI with no first \
159             turn and it stalls headless: {args:?}"
160        );
161        assert!(
162            args.windows(2)
163                .any(|w| w[0] == "--output-format" && w[1] == "stream-json"),
164            "the OUTPUT format is what makes the capture a JSONL event stream \
165             the Layer 1 stream parser can read: {args:?}"
166        );
167        assert!(args.iter().any(|a| a == "--dangerously-skip-permissions"));
168        assert!(
169            !args.iter().any(|arg| arg.contains("DEVFLOW_RESULT")),
170            "no positional prompt: the initial user turn travels on stdin, \
171             written by the monitor: {args:?}"
172        );
173    }
174
175    #[test]
176    fn codex_wraps_prompt_in_exec_and_json() {
177        let prompt = stage_prompt(Stage::Code, 7);
178        let (program, args) = adapter_for(AgentKind::Codex).exec_command(7, &prompt, &[]);
179        assert_eq!(program, "codex");
180        let joined = args.join(" ");
181        assert!(joined.contains("exec"));
182        assert!(joined.contains("--sandbox workspace-write"));
183        assert!(joined.contains("--json"));
184    }
185
186    #[test]
187    fn opencode_wraps_prompt_in_run() {
188        let prompt = stage_prompt(Stage::Code, 7);
189        let (program, args) = adapter_for(AgentKind::OpenCode).exec_command(7, &prompt, &[]);
190        assert_eq!(program, "opencode");
191        assert_eq!(args, ["run", prompt.as_str()]);
192    }
193
194    /// 13-06 dogfood regression (Codex leg): linked-worktree git metadata
195    /// lives under the main repo's `.git/` — outside the workspace-write
196    /// sandbox — and Codex read-only-mounts the cwd's resolved git dir, so
197    /// BOTH the common `.git` and the worktree admin dir must be granted
198    /// (verified with `codex sandbox` probes). Without roots, no override.
199    #[test]
200    fn codex_grants_writable_roots_for_worktree_git_metadata() {
201        let prompt = stage_prompt(Stage::Code, 7);
202        let roots = vec![
203            PathBuf::from("/repo/.git"),
204            PathBuf::from("/repo/.git/worktrees/phase-07"),
205        ];
206        let (_, args) = adapter_for(AgentKind::Codex).exec_command(7, &prompt, &roots);
207        let joined = args.join(" ");
208        assert!(
209            joined.contains(
210                r#"-c sandbox_workspace_write.writable_roots=["/repo/.git","/repo/.git/worktrees/phase-07"]"#
211            ),
212            "codex must whitelist the common .git AND the worktree admin dir: {joined}"
213        );
214
215        let (_, args) = adapter_for(AgentKind::Codex).exec_command(7, &prompt, &[]);
216        assert!(
217            !args.join(" ").contains("writable_roots"),
218            "no override without an extra root"
219        );
220    }
221
222    /// 13-06 dogfood regression: signed commits fail inside the Codex
223    /// sandbox (no route to the operator's signing agent) — codex scopes an
224    /// unsigned-commit override to its own process tree via GIT_CONFIG_*
225    /// env; agents without a sandbox get no extra env.
226    #[test]
227    fn codex_disables_signing_via_env_others_do_not() {
228        let env = adapter_for(AgentKind::Codex).extra_env();
229        assert!(env.contains(&("GIT_CONFIG_KEY_0".into(), "commit.gpgsign".into())));
230        assert!(env.contains(&("GIT_CONFIG_KEY_1".into(), "tag.gpgsign".into())));
231        assert!(adapter_for(AgentKind::Claude).extra_env().is_empty());
232        assert!(adapter_for(AgentKind::OpenCode).extra_env().is_empty());
233    }
234
235    /// D-13: `preflight`'s default body is `Ok(())` for every built-in
236    /// adapter — none of Claude/Codex/OpenCode override it in Phase 17 (no
237    /// reviewer-set storage exists yet in `state.rs`/`config.rs`, review
238    /// consensus #6).
239    #[test]
240    fn default_preflight_is_ok_for_built_in_adapters() {
241        let state = crate::state::State::new(
242            1,
243            AgentKind::Claude,
244            crate::mode::Mode::Auto,
245            PathBuf::from("/repo"),
246        );
247        assert!(adapter_for(AgentKind::Claude).preflight(&state).is_ok());
248        assert!(adapter_for(AgentKind::Codex).preflight(&state).is_ok());
249        assert!(adapter_for(AgentKind::OpenCode).preflight(&state).is_ok());
250    }
251}
devflow_core/agents/mod.rs

devflow_core/agents/
mod.rs
