You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The flagship workflow. `agentic-harness code` inspects the repo, reads `AGENTS.md` / `CLAUDE.md`, captures git diff context, drafts a plan, hands the brief to your installed coding LLM, then runs detected checks (`cargo test`, etc.) and optionally commits or opens a PR. No prompt? It defaults to a sensible "smallest safe step" brief.
90
+
The flagship workflow. `agentic-harness start` opens the guided TUI front door;
91
+
`agentic-harness code` is the direct automation path. The coding loop inspects
92
+
the repo, reads `AGENTS.md` / `CLAUDE.md`, captures git diff context, drafts a
93
+
plan, hands the brief to your installed coding LLM, runs detected checks
94
+
(`cargo test`, etc.), and optionally commits or opens a PR. No prompt? It
95
+
defaults to a sensible "smallest safe step" brief.
89
96
90
97
```bash
91
98
# Detects whichever of claude / codex / cursor / wind-server you have
92
99
agentic-harness code --workspace . --llm auto \
93
100
--prompt "Add a flag to skip the network call in test mode" \
101
+
--deny-path .env \
102
+
--approve-dependencies \
94
103
--commit "feat: --offline flag" \
95
104
--pr
96
105
```
97
106
98
-
The harness writes a run-scoped brief to `.agentic-harness/runs/<id>/coding-brief.md`, streams progress, captures the agent result, and saves a structured summary at `.agentic-harness/runs/latest.{md,json}` so a person or another agent can read what happened.
107
+
The harness writes a run-scoped brief to
108
+
`.agentic-harness/runs/<id>/coding-brief.md`, streams progress, captures the
109
+
agent result, and saves both latest summaries and a durable bundle under
0 commit comments