deepflow 0.1.53 → 0.1.55

package/README.md

@@ -75,25 +75,40 @@ claude
 You write the specs, then walk away. The AI runs the full pipeline — hypothesis generation, parallel spikes, implementation, adversarial self-selection, verification — without any human intervention.
 
 ```bash
-# You define WHAT (the specs already exist as doing-*)
-# The AI figures out HOW, overnight
+# You define WHAT (the specs), the AI figures out HOW, overnight
 
-deepflow auto                    # process all doing-* specs
+deepflow auto                    # process all specs in specs/
 deepflow auto --parallel=3       # 3 approaches in parallel
 deepflow auto --hypotheses=4     # 4 hypotheses per cycle
 deepflow auto --max-cycles=5     # cap retry cycles
 ```
 
 **What the AI does alone:**
-1. Reads each `doing-*` spec
-2. Generates N hypotheses for how to implement it
+1. Discovers specs (auto-promotes plain specs to `doing-*`)
+2. Generates N hypotheses for how to implement each spec
 3. Runs parallel spikes in isolated worktrees (one per hypothesis)
 4. Implements the passing approaches
 5. Adversarial selection: a fresh AI context compares approaches by artifacts only (never reads code), picks the best or rejects all
 6. If rejected: generates new hypotheses, retries (up to max-cycles)
 7. On convergence: verifies, merges to main
 
-**What you do:** Write specs (via interactive mode or manually), run `deepflow auto`, read the morning report at `.deepflow/auto-report.md`.
+**What you do:** Write specs (via interactive mode or manually) in `specs/`, run `deepflow auto`, read the morning report at `.deepflow/auto-report.md`. No need to run `/df:plan` first — auto mode promotes plain specs to `doing-*` automatically.
+
+**How to use:**
+```bash
+# In Claude Code — create and approve a spec
+$ claude
+> /df:discover auth
+> /df:spec auth          # creates specs/auth.md
+> /exit
+
+# In your terminal — run auto mode and walk away
+$ deepflow auto
+
+# Next morning — check what happened
+$ cat .deepflow/auto-report.md
+$ git log --oneline
+```
 
 **Safety:** Never pushes to remote. Failed approaches recorded in `.deepflow/experiments/` and never repeated. Specs validated before processing (malformed specs are skipped).
 
@@ -109,7 +124,7 @@ deepflow auto --max-cycles=5     # cap retry cycles
                                     Merge or retry
  Read morning report
  ─────────────────────────────────  ──────────────────────────────────
-         specs/doing-*.md is the handoff point
+         specs/*.md is the handoff point
 ```
 
 ## The Flow (Interactive)
@@ -150,6 +165,7 @@ deepflow auto --max-cycles=5     # cap retry cycles
 
 ```
 deepflow auto
+    | Discover specs (auto-promote plain specs to doing-*)
     | For each doing-* spec:
     |
     |   Validate spec (malformed? skip)

package/bin/deepflow-auto.sh

@@ -114,14 +114,30 @@ discover_specs() {
     exit 1
   fi
 
-  # Collect doing-*.md files (read-only — we never modify these)
+  # Collect doing-*.md files
   for f in "${specs_dir}"/doing-*.md; do
     [[ -e "$f" ]] || continue
     found+=("$f")
   done
 
+  # Auto-promote plain specs (not doing-*, done-*, or .debate-*) to doing-*
+  for f in "${specs_dir}"/*.md; do
+    [[ -e "$f" ]] || continue
+    local base
+    base="$(basename "$f")"
+    # Skip already-prefixed and auxiliary files
+    case "$base" in
+      doing-*|done-*|.debate-*|.*) continue ;;
+    esac
+    local new_path="${specs_dir}/doing-${base}"
+    mv "$f" "$new_path"
+    auto_log "Auto-promoted spec: ${base} -> doing-${base}"
+    echo "Promoted: ${base} -> doing-${base}" >&2
+    found+=("$new_path")
+  done
+
   if [[ ${#found[@]} -eq 0 ]]; then
-    echo "Error: no specs/doing-*.md files found in ${specs_dir}" >&2
+    echo "Error: no specs found in ${specs_dir}" >&2
     exit 1
   fi
 
@@ -420,6 +436,7 @@ run_single_spike() {
   local slug="$2"
   local hypothesis="$3"
   local method="$4"
+  local spec_file="$5"
 
   local worktree_path="${PROJECT_ROOT}/.deepflow/worktrees/${spec_name}-${slug}"
   local branch_name="df/${spec_name}-${slug}"
@@ -439,6 +456,12 @@ run_single_spike() {
     }
   fi
 
+  # Extract acceptance criteria from spec (the human's judgment proxy)
+  local acceptance_criteria=""
+  if [[ -f "$spec_file" ]]; then
+    acceptance_criteria="$(sed -n '/^## Acceptance Criteria/,/^## /{ /^## Acceptance Criteria/d; /^## /d; p; }' "$spec_file")"
+  fi
+
   # Build spike prompt
   local spike_prompt
   spike_prompt="You are running a spike experiment to validate a hypothesis for spec '${spec_name}'.
@@ -449,13 +472,19 @@ Hypothesis: ${hypothesis}
 Method: ${method}
 --- END HYPOTHESIS ---
 
+--- ACCEPTANCE CRITERIA (from spec — the human's judgment proxy) ---
+${acceptance_criteria}
+--- END ACCEPTANCE CRITERIA ---
+
 Your tasks:
 1. Validate this hypothesis by implementing the minimum necessary to prove or disprove it.
+   The spike must demonstrate that the approach can satisfy the acceptance criteria above.
 2. Write an experiment file at: .deepflow/experiments/${spec_name}--${slug}--active.md
    The experiment file should contain:
    - ## Hypothesis: restate the hypothesis
    - ## Method: what you did to validate
    - ## Results: what you observed
+   - ## Criteria Check: for each acceptance criterion, can this approach satisfy it? (yes/no/unclear)
    - ## Conclusion: PASSED or FAILED with reasoning
 3. Write a result YAML file at: .deepflow/results/spike-${slug}.yaml
    The YAML must contain:
@@ -557,7 +586,7 @@ run_spikes() {
     auto_log "Spawning spike for ${slug} (hypothesis ${i}/${count})"
     echo "Spawning spike: ${slug}"
 
-    run_single_spike "$spec_name" "$slug" "$hypothesis" "$method" &
+    run_single_spike "$spec_name" "$slug" "$hypothesis" "$method" "$spec_file" &
     pids+=($!)
   done
 
@@ -859,6 +888,13 @@ $(cat "$experiment_file")
   # -----------------------------------------------------------------------
   # 2. Build selection prompt
   # -----------------------------------------------------------------------
+
+  # Extract acceptance criteria from spec (the human's judgment proxy)
+  local acceptance_criteria=""
+  if [[ -f "$spec_file" ]]; then
+    acceptance_criteria="$(sed -n '/^## Acceptance Criteria/,/^## /{ /^## Acceptance Criteria/d; /^## /d; p; }' "$spec_file")"
+  fi
+
   local selection_prompt
   selection_prompt="You are an adversarial quality judge in an autonomous development workflow.
 Your job is to compare implementation approaches for spec '${spec_name}' and select the best one — or reject all if quality is insufficient.
@@ -867,6 +903,11 @@ IMPORTANT:
 - This selection phase ALWAYS runs, even with only 1 approach. With a single approach you act as a quality gate.
 - You CAN and SHOULD reject all approaches if the quality is insufficient. Do not rubber-stamp poor work.
 - Base your judgment ONLY on the artifacts provided below. Do NOT read code files.
+- Judge each approach against the ACCEPTANCE CRITERIA below — these represent the human's intent.
+
+--- ACCEPTANCE CRITERIA (from spec) ---
+${acceptance_criteria}
+--- END ACCEPTANCE CRITERIA ---
 
 There are ${#approach_slugs[@]} approach(es) to evaluate:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "deepflow",
-  "version": "0.1.53",
+  "version": "0.1.55",
   "description": "Stay in flow state - lightweight spec-driven task orchestration for Claude Code",
   "keywords": [
     "claude",