Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/conorbronsdon/agent-memory-kit/dream-applygit clone --depth 1 https://github.com/conorbronsdon/agent-memory-kitWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00020 | $0.02101 |
| Opus 5 | $0.00010 | $0.01051 |
| Sonnet 5 | $0.00004 | $0.00420 |
| Haiku 4.5 | $0.00002 | $0.00210 |
Grade A, and why
dream-apply scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/dream-apply — review + apply a curator pass
Background: docs/curation.md. Nothing is auto-applied — every proposal goes through review.
Usage
/dream-apply {ISO-timestamp}
/dream-apply latest # most recent .dreams/ subdir
1. Resolve the dream dir
Resolve $MEMORY_DIR (same order as /dream). DREAMS_ROOT="$MEMORY_DIR/.dreams". If $ARGUMENTS is latest/empty, pick the most recent subdir (ISO names sort lexically). Else treat $ARGUMENTS as the timestamp. If the dir doesn't exist, list available dreams and stop.
2. Load the artifact
Read proposals.json and REPORT.md. If either is missing, stop with an error.
3. Show the report header
Print the header + counts only — don't dump the whole report.
4. Walk each proposal
For each:
a. Print id, action, target, confidence, reasoning, evidence, and the current → proposed excerpts.
b. Ask via AskUserQuestion: Accept / Reject / Edit then accept / Skip rest.
highconfidence → order Accept-first.medium→ order Reject-first (forces a read).flag→ order Reject-first; Accept is opt-in only.
c. On Accept:
modify→ Editmemory/{target}, replacingcurrent_excerptwithproposed_excerpt.archive→ all five steps. An archive that stops early leaves the file reading as live:- Decide from the filesystem first, then the row. Run all three, before anything else:
test -f "$MEMORY_DIR/archive/{target}" # A: already under archive/ test -f "$MEMORY_DIR/{target}" # B: still live in the root { grep -cF "](archive/{target})" "$MEMORY_DIR/ARCHIVE.md" 2>/dev/null || echo 0; } | head -1 # C: tombstone rowsA B C Do hit — any Genuinely retired. Stop. hit hit any Two copies exist. Stop and surface — the root copy is live and an archived one already exists, so which survives is a human's call. — — any Target is gone (deleted or renamed between dream and apply). Stop and surface. — hit 0 Ordinary archive. All five steps. — hit >0 Crashed earlier run. Resume: skip step 2 — appending again is what produces the duplicate rows lint check 9 reports. Keep the earliest row and drop the rest; that date is the true retirement date, so stamp with it instead of today. Do step 3 only if grep -c '^archived:' "$MEMORY_DIR/{target}"is 0. Then 4 and 5.
- Decide from the filesystem first, then the row. Run all three, before anything else:
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 133 lines · 20 tokens per session scan A 9e4f5ffeeef6
dream-apply is a command published in the GitHub repository conorbronsdon/agent-memory-kit (2 stars, last pushed 28d ago), licensed MIT. It adds 20 tokens to every session and 2,101 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
create-pr
Create a pull request for the current branch against main.
create-branch
Create a new git branch with an associated worktree following the project naming convention.
list-branches
List project branches with their worktree status.
finalize
Finalize the current work: format, test, fix, commit, and push.
commit
Create a new commit for all of our uncommitted changes run git status && git diff HEAD && git status --porcelain to see what files are uncommitted add the untracked and changed files.
capture
Record what this session learned so a future session in a related repository can find it.