[week-05] 25622005 by psyoung82 · Pull Request #265 · Q00/ai-agent-engineering-101

lab(week-05): same question over stdio and HTTP, both read_file + calculator -> 69504, host code unchanged

psyoung82

and

claude

committed

bd22942

lab(week-05): Claude Code as second client on the same HTTP server, read_file + calculator -> 69504

psyoung82

and

claude

committed

be885af

lab(week-05): add clock to the server only; unchanged host shows 3 tools and calls it over stdio and HTTP

psyoung82

and

claude

committed

9eb6c9a

feat(week-05): scenarios.json (week-04's four, limits fixed before any run) and condition-independent prompts

psyoung82

and

claude

committed

ad65f77

feat(week-05): market MCP server with bearer-token parties, turn checks, token-limit refusals, and the buyer-side injection

psyoung82

and

claude

committed

aeda9e4

feat(week-05): auth_checks.txt from four live checks: 401 without token, foreign negotiation, out of turn, outside token limit

psyoung82

and

claude

committed

e588a70

feat(week-05): host (one turn, bearer header), runner (episodes, results.csv, resume), pass-turn admin route, offline replay test

psyoung82

and

claude

committed

db13688

run(week-05): smoke episode, server_inject scenario 2 run 1, deal at 38, buyer named the injection and ignored it

psyoung82

and

claude

committed

7ede86c

run(week-05): run 1 prompt_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios left open at 8 moves

psyoung82

and

claude

committed

7a73733

run(week-05): run 1 server_inject complete, 4 episodes: 2 deals inside limits, 2 impossible scenarios open, no refusals

psyoung82

and

claude

committed

37b7265

run(week-05): run 2 prompt_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open

psyoung82

and

claude

committed

5be5a7f

run(week-05): run 2 server_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open, no refusals

psyoung82

and

claude

committed

17227ba

run(week-05): run 3 prompt_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open

psyoung82

and

claude

committed

cf40035

run(week-05): run 3 server_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open; all 24 required episodes done

psyoung82

and

claude

committed

007380e

docs(week-05): REPORT.md draft, four parts: setup, results, FIPA-ACL vs market table, interpretation with log lines

psyoung82

and

claude

committed

5e8c6e0

docs(week-05): match the lecture's reference-run layout (week-04 comparison table, labelled log findings, log excerpt), rewrite the interpretation paragraph

psyoung82

and

claude

committed

d0f02ab

docs(week-05): restructure the interpretation paragraph: claim first, three numbered grounds, two residual findings

psyoung82

and

claude

committed

0d3c9aa

docs(week-05): state the negative result on the core question and what the run did not show

psyoung82

and

claude

committed

4e05453

docs(week-05): interpretation as one passage with three bullet lists (grounds, what was not shown, model-side findings)

psyoung82

and

claude

committed

28dec0f

misc: add stderr logs

psyoung82

committed

3735c6c

Navigation Menu

Repository navigation

[week-05] 25622005 - #265

[week-05] 25622005#265
psyoung82 wants to merge 24 commits into
Q00:mainfrom
psyoung82:week-05

Commits on Sep 29, 2026

feat(week-05): move week-01 calculator and read_file into an MCP server (tools_server.py)

lab(week-05): call tools/list by hand over HTTP; missing Mcp-Method or clientCapabilities gets 400

feat(week-05): turn the week-01 loop into an MCP host; tools from tools/list, execution via tools/call

wip(week-05): stdio host run reached the model but failed with 402 insufficient_quota on the Anthropic key

Commits on Oct 5, 2026

lab(week-05): same question over stdio and HTTP, both read_file + calculator -> 69504, host code unchanged

lab(week-05): Claude Code as second client on the same HTTP server, read_file + calculator -> 69504

lab(week-05): add clock to the server only; unchanged host shows 3 tools and calls it over stdio and HTTP

feat(week-05): scenarios.json (week-04's four, limits fixed before any run) and condition-independent prompts

feat(week-05): market MCP server with bearer-token parties, turn checks, token-limit refusals, and the buyer-side injection

feat(week-05): auth_checks.txt from four live checks: 401 without token, foreign negotiation, out of turn, outside token limit

feat(week-05): host (one turn, bearer header), runner (episodes, results.csv, resume), pass-turn admin route, offline replay test

run(week-05): smoke episode, server_inject scenario 2 run 1, deal at 38, buyer named the injection and ignored it

run(week-05): run 1 prompt_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios left open at 8 moves

run(week-05): run 1 server_inject complete, 4 episodes: 2 deals inside limits, 2 impossible scenarios open, no refusals

run(week-05): run 2 prompt_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open

run(week-05): run 2 server_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open, no refusals

run(week-05): run 3 prompt_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open

run(week-05): run 3 server_inject, 4 episodes: 2 deals inside limits, 2 impossible scenarios open; all 24 required episodes done

docs(week-05): REPORT.md draft, four parts: setup, results, FIPA-ACL vs market table, interpretation with log lines

docs(week-05): match the lecture's reference-run layout (week-04 comparison table, labelled log findings, log excerpt), rewrite the interpretation paragraph

docs(week-05): restructure the interpretation paragraph: claim first, three numbered grounds, two residual findings

docs(week-05): state the negative result on the core question and what the run did not show

docs(week-05): interpretation as one passage with three bullet lists (grounds, what was not shown, model-side findings)

misc: add stderr logs