fix: Codex review blockers — nightly prompt schema, API docs, task enum, ADR-003

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Paul Kyle

ADR-003-memory-harness-boundary.md

@@ -27,7 +27,7 @@ This looks correct. It is not. It is **save and pray** — the implicit assumpti
 
 The failure mode is always the same: **the agent doesn't know what it doesn't know.** It proceeds with confidence on incomplete context. There is no error. There is no warning. The dashboard is green. The memories are gone.
 
-This is not a theoretical risk. Research from 2026 shows agents completing tasks while losing access to 87% of stored knowledge under complexity. Task completion rates stay high while memory recall drops to 13%. The agent finishes the work. The work is wrong.
+This is not a theoretical risk. Early benchmarks on agent memory retrieval suggest that recall degrades significantly under task complexity — agents complete work while missing large portions of their stored knowledge. Task completion metrics stay green even as retrieval quality degrades. The agent finishes the work. The work is wrong.
 
 ## Architecture: Two Systems, One Contract
 

README.md

@@ -76,7 +76,7 @@ pip install -e .
 # Create your memory directory
 mkdir -p ~/.palinode/{people,projects,decisions,insights,daily}
 cd ~/.palinode && git init
-cp ~/palinode/palinode.config.yaml.example palinode.config.yaml
+cp /path/to/palinode/palinode.config.yaml.example palinode.config.yaml  # adjust path
 
 # Start services
 PALINODE_DIR=~/.palinode palinode-api        # REST API on :6340
@@ -213,11 +213,11 @@ Optional: a chat model for consolidation (any 7B+ works), OpenClaw for agent plu
 | `GET/POST` | `/triggers` | Prospective recall triggers |
 | `POST` | `/consolidate` | Run or preview compaction |
 | `GET` | `/list` | Browse files by type |
-| `GET` | `/read/{path}` | Read a memory file |
-| `GET` | `/history/{path}` | Git log for a file |
+| `GET` | `/read?file_path=...` | Read a memory file |
+| `GET` | `/history/{file_path}` | Git log for a file |
 | `GET` | `/diff` | Recent changes |
-| `GET` | `/blame` | Git blame |
-| `GET` | `/timeline` | File evolution |
+| `GET` | `/blame/{file_path}` | Git blame |
+| `GET` | `/timeline/{file_path}` | File evolution |
 | `POST` | `/rollback` | Revert a file |
 | `POST` | `/push` | Push to git remote |
 | `POST` | `/reindex` | Rebuild indices |

docs/OBSIDIAN-SETUP.md

@@ -12,4 +12,4 @@ Because Palinode stores all memories as standard markdown files, you can use [Ob
 
 - **Visual Graph:** See how your agent connects people, projects, and ideas (frontmatter `entities` arrays act as tags/links if you configure Obsidian's Dataview)
 - **Manual Curation:** If the agent gets a detail wrong, just type to fix it. The file watcher will pick up your changes within 2 seconds.
-- **GrueBrain Rationalization:** If you use GrueBrain, you can symlink specific subdirectories (like `projects/` or `daily/`) into your main GrueBrain vault.
+- **Second Brain Integration:** You can symlink specific subdirectories (like `projects/` or `daily/`) into an existing Obsidian vault.

palinode/cli/consolidate.py

@@ -7,7 +7,7 @@
 @click.option("--dry-run", is_flag=True, help="Preview changes without applying")
 @click.option("--format", "fmt", type=click.Choice(["json", "text"]), help="Output format")
 def consolidate(nightly, dry_run, fmt):
-    """Run or preview weekly compaction."""
+    """Run or preview memory compaction (weekly full or --nightly lightweight)."""
     try:
         data = api_client.consolidate(dry_run=dry_run, nightly=nightly)
 

palinode/cli/prompt.py

@@ -20,7 +20,7 @@ def prompt():
 
 
 @prompt.command(name="list")
-@click.option("--task", type=click.Choice(["compaction", "extraction", "update", "classification"]),
+@click.option("--task", type=click.Choice(["compaction", "extraction", "update", "classification", "nightly-consolidation"]),
               help="Filter by task type")
 @click.option("--format", "fmt", type=click.Choice(["text", "json"]), default=None)
 def prompt_list(task: str | None, fmt: str | None) -> None:

palinode/mcp.py

@@ -479,7 +479,7 @@ async def list_tools() -> list[types.Tool]:
                     "task": {
                         "type": "string",
                         "description": "For 'list': filter by task type",
-                        "enum": ["compaction", "extraction", "update", "classification"],
+                        "enum": ["compaction", "extraction", "update", "classification", "nightly-consolidation"],
                     },
                 },
                 "required": ["action"],

specs/prompts/nightly-consolidation.md

@@ -19,11 +19,11 @@ For each project mentioned in the daily notes, propose UPDATE or SUPERSEDE opera
 ## Rules
 - Only UPDATE (append new info) or SUPERSEDE (replace outdated line with current info)
 - Do NOT ARCHIVE, MERGE, or KEEP — those are weekly operations
-- Each operation targets a specific fact_id in the status file
+- Each operation targets a specific `id` (fact ID) in the status file
 - If a status line is now outdated by today's work, SUPERSEDE it
 - If today adds new information, UPDATE with a new status line
 - Be concise: one line per status entry, format: `- [YYYY-MM-DD] summary`
 
 ## Output
 Return a JSON array of operations:
-[{"op": "UPDATE", "fact_id": "...", "content": "..."}, ...]
+[{"op": "UPDATE", "id": "...", "new_text": "..."}, ...]